1 const char jcc_rcs[] = "$Id: jcc.c,v 1.195 2008/10/13 16:04:37 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 1.195 2008/10/13 16:04:37 fabiankeil
37 * Make sure we don't try to reuse tainted server sockets.
39 * Revision 1.194 2008/10/12 18:35:18 fabiankeil
40 * The last commit was a bit too ambitious, apparently the content
41 * length adjustment is only necessary if we aren't buffering.
43 * Revision 1.193 2008/10/12 15:57:35 fabiankeil
44 * Fix content length calculation if we read headers
45 * and the start of the body at once. Now that we have
46 * FEATURE_CONNECTION_KEEP_ALIVE, it actually matters.
48 * Revision 1.192 2008/10/11 18:19:14 fabiankeil
49 * Even more chat() cosmetics.
51 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
52 * Reformat some comments in chat().
54 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
55 * In case of chunk-encoded content, stop reading if
56 * the buffer looks like it ends with the last chunk.
58 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
59 * Let server_response_is_complete() deal properly with
60 * content that is neither buffered nor read all at once.
62 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
63 * Flush work-in-progress changes to keep outgoing connections
64 * alive where possible. Incomplete and mostly #ifdef'd out.
66 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
67 * Add mutex lock support for _WIN32.
69 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
70 * Prepare for critical sections on Windows by adding a
71 * layer of indirection before the pthread mutex functions.
73 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
74 * Remove FEATURE_COOKIE_JAR.
76 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
77 * - Silence LLVM/Clang complaint.
78 * - Make received_hup_signal static.
79 * - Hide definitions for basedir, pidfile and received_hup_signal
80 * from __EMX__ as they only seem to be used in case of #ifdef unix.
82 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
83 * Accept Shoutcast responses again. Problem reported
84 * and fix suggested by Stefan in #2062860.
86 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
87 * Fix possible NULL-pointer dereference reported
88 * by din_a4 in #2003937. Pointy hat to me.
90 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
91 * Streamline sed()'s prototype and declare
92 * the header parse and add structures static.
94 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
95 * - Mark csp as immutable for send_crunch_response().
96 * - Fix comment spelling.
98 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
99 * Factor update_server_headers() out of sed(), ditch the
100 * first_run hack and make server_patterns_light static.
102 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
103 * Don't provide get_header() with the whole client state
104 * structure when it only needs access to csp->iob.
106 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
107 * Make the "read the rest of the headers" loop a bit more readable.
109 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
110 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
111 * - Remove useless comment.
113 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
114 * Fix comment grammar.
116 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
117 * Remove the pointless buffer in client_protocol_is_unsupported().
119 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
120 * Least-effort fix for bug #1821930 (reported by Lee):
121 * If the response doesn't look like HTTP,
122 * tell the client and log the problem.
124 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
125 * Don't pass the whole csp structure to flush_socket()
126 * when it only needs a file descriptor and a buffer.
128 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
129 * Remove kill-popups action.
131 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
132 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
134 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
135 * Remove the treat-forbidden-connects-like-blocks action. We now
136 * use the "blocked" page for forbidden CONNECT requests by default.
138 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
139 * Also use shiny new connect_port_is_forbidden() in jcc.c.
141 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
142 * Rename url_actions() to get_url_actions() and let it
143 * use the standard parameter ordering.
145 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
146 * Let forward_url() use the standard parameter ordering
147 * and mark its second parameter immutable.
149 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
150 * Remove the "Listening ... for local connections only" log message.
151 * Whether or not remote connections are able to reach Privoxy is up
152 * to the operating system.
154 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
155 * Prevent the log messages for CONNECT requests to unacceptable
156 * ports from printing the limit-connect argument as [null] if
157 * limit-connect hasn't been explicitly enabled.
159 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
160 * Make sure all console-mode apps get a usage() instance
162 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
163 * Reword NO_SERVER_DATA_RESPONSE to make it harder
164 * to misunderstand what the message is all about.
166 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
167 * Unbreak trustfile which previously didn't work without
168 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
170 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
171 * Plug memory leak. Spotted by Valgrind, triggered by
172 * Privoxy-Regression-Test feeding proxyfuzz.py.
174 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
175 * In the HTTP snipplets, refer to the client as client.
177 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
178 * Emit a log message when activating the MS IIS5 hack.
180 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
181 * Log the "weak randomization factor" warning only
182 * once for mingw32 and provide some more details.
184 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
185 * Initialize log module after initializing mutexes, future
186 * deadlocks in that code should now work cross-platform.
188 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
189 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
190 * Reported by Bert van Leeuwen in BR#1818808.
192 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
193 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
195 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
196 * When in daemon mode, close stderr after the configuration file has been
197 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
199 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
200 * - Fix a crash when parsing invalid requests whose first header
201 * is rejected by get_header(). Regression (re?)introduced
202 * in r1.143 by yours truly.
203 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
204 * client_cookie_adder() to make sure send-vanilla-wafer can be
205 * controlled through tags (and thus regression-tested).
207 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
208 * - Move get_filter_function() from jcc.c to filters.c
209 * so the filter functions can be static.
210 * - Don't bother filtering body-less responses.
212 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
213 * Execute content filters through execute_content_filter().
215 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
216 * Initialize req to NULL to make sure it's defined if the
217 * first read_socket() call fails. Reported by icmp30.
219 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
220 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
221 * extensive comments moved to user manual.
223 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
224 * Don't crash if a broken header filter wiped out the request line.
226 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
227 * Fix byte_count calculation in case of flushes
228 * and don't parse the server headers a second time.
230 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
231 * - If there's a connection problem after we already forwarded
232 * parts of the original content, just hang up. Fixes BR#1776724.
233 * - Fix warnings about unused code on mingw32.
234 * - In case of flushes, calculate the byte count
235 * less incorrectly (I think).
237 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
238 * Add some more prototypes for static functions.
240 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
241 * Comment out request_contains_null_bytes() until it's used again.
243 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
244 * #1763173 from Stefan Huehner: s@const static@static const@
245 * and declare some more functions static.
247 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
248 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
249 * and explain why they were rejected in the first place.
250 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
251 * CONNECT requests. The request line was missing.
252 * - Add two more XXX reminders as we don't have enough already.
254 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
255 * As Hal noticed, checking dispatch_cgi() as the last cruncher
256 * looks like a bug if CGI requests are blocked unintentionally,
257 * so don't do it unless the user enabled the new config option
258 * "allow-cgi-request-crunching".
260 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
261 * - Allow to rewrite the request destination behind the client's back.
262 * - Turn the weird-looking unconditional for loop that
263 * reads the client request into a conditional while loop.
264 * Move the stuff that only runs once out of the loop.
265 * - Move parts of chat(), server_content_type() and the
266 * necessary stuff to fix BR#1750917 into get_filter_function().
268 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
269 * Temporary workaround for BR#1730105.
271 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
272 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
273 * deadlocks and crashes on OpenBSD and possibly other OS with neither
274 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
275 * Thanks to Ralf Horstmann for report and solution.
277 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
278 * Add forward-override{} to change the forwarding settings through
279 * action sections. This is mainly interesting to forward different
280 * clients differently (for example based on User-Agent or request
283 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
284 * - Let usage() mention the --chroot parameter.
285 * - Use read_socket() consistently and always leave
286 * the last buffer byte alone, even in cases where
287 * null termination (currently) doesn't matter.
289 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
290 * - Fix config file loading on Unix if no config file is specified.
291 * Since r1.97 Privoxy would always interpret the last argument as
292 * config file, even if it's a valid command line option.
293 * - Abort in case of unrecognized command line options. Closes #1719696.
294 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
295 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
297 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
298 * - Don't rerun crunchers that only depend on the request URL.
299 * - Don't count redirects and CGI requests as "blocked requests".
301 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
302 * Support crunching based on tags created by server-header taggers.
304 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
305 * Make HTTP snippets static (again). Add a Content-Type for those
306 * with content so the browser doesn't guess it based on the URL.
308 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
309 * Move crunching and request line rebuilding out of chat().
311 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
312 * Introduce tags as alternative way to specify which
313 * actions apply to a request. At the moment tags can be
314 * created based on client and server headers.
316 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
317 * Don't CLF-log CONNECT requests twice.
319 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
320 * Log the source address for ACL-related connection drops.
322 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
323 * New config option: enforce-blocks.
325 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
326 * - Move null byte check into separate function.
327 * - Don't confuse the client with error pages
328 * if a CONNECT request was already confirmed.
330 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
331 * Speed up NULL byte escaping and only log the complete
332 * NULL byte requests with header debugging enabled.
334 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
335 * Answer requests that contain NULL bytes with
336 * a custom response instead of waiting for more
337 * data until the client eventually hangs up.
339 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
340 * - Move delivery and logging of crunched responses
341 * from chat() into send_crunch_response().
342 * - Display the reason for generating http_responses.
343 * - Log the content length for LOG_LEVEL_CLF correctly
344 * (still incorrect for some fixed responses).
345 * - Reword an incorrect comment about
346 * treat-forbidden-connects-like-blocks violating
348 * - Add some log messages.
350 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
351 * Move mutex initialization into separate
352 * function and exit in case of errors.
354 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
355 * - Start to reduce chat()'s line count and move
356 * parts of it into separate functions.
357 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
359 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
360 * - Add Proxy-Agent header to HTTP snippets that are
361 * supposed to reach HTTP clients only.
362 * - Made a few CONNECT log messages more descriptive.
363 * - Catch completely empty server responses (as seen
364 * with Tor's fake ".noconnect" top level domain).
365 * - Use shiny new "forwarding-failed" template for socks errors.
367 * Revision 1.118 2007/01/07 07:43:43 joergs
368 * AmigaOS4 support added.
370 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
371 * Added config option accept-intercepted-requests
372 * and disabled it by default.
374 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
375 * Reverted parts of my last commit
376 * to keep error handling working.
378 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
379 * Fixed gcc43 conversion warnings.
381 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
382 * Fix -pedantic ISO C warning about converting
383 * from function pointer to object pointer.
385 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
386 * Silence compiler warning I introduced with my last commit.
388 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
389 * Mutex protect rand() if POSIX threading
390 * is used, warn the user if that's not possible
391 * and stop using it on _WIN32 where it could
394 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
395 * Don't prevent core dumps by catching SIGABRT.
396 * It's rude and makes debugging unreasonable painful.
398 * Revision 1.110 2006/12/13 14:52:53 etresoft
399 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
401 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
402 * Privoxy is now able to run as intercepting
403 * proxy in combination with any packet filter
404 * that does the port redirection. The destination
405 * is extracted from the "Host:" header which
406 * should be available for nearly all requests.
408 * Moved HTTP snipplets into jcc.c.
409 * Added error message for gopher proxy requests.
411 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
412 * Only unlink the pidfile if it's actually used.
414 * Change order of interception checks to make
415 * it possible to block or redirect requests for
418 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
419 * Make pthread mutex locking more generic. Instead of
420 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
421 * and use mutex locking unless there is an _r function
422 * available. Better safe than sorry.
424 * Fixes "./configure --disable-pthread" and should result
425 * in less threading-related problems on pthread-using platforms,
426 * but it still doesn't fix BR#1122404.
428 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
429 * Move pthread.h inclusion from jcc.c to jcc.h.
430 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
432 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
433 * Don't exit after receiving the second SIGHUP on Solaris.
435 * Fixes BR 1052235, but the same problem may exist on other
436 * systems. Once 3.0.6 is out we should use sigset()
437 * where available and see if it breaks anything.
439 * Revision 1.104 2006/09/23 13:26:38 roro
440 * Replace TABs by spaces in source code.
442 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
443 * Fix +redirect{}. Didn't work with -fast-redirects.
445 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
446 * Respond with 400 and a short text message
447 * if the client tries to use Privoxy as FTP proxy.
449 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
450 * Make number of retries in case of forwarded-connect problems
451 * a config file option (forwarded-connect-retries) and use 0 as
454 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
455 * Set random(3) seed.
457 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
458 * Follow the OpenBSD port's lead and protect the resolve
459 * functions on OpenBSD as well.
461 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
462 * --user fix. Only use the user as group if no group is specified.
463 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
465 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
466 * Windows service (re-)integration
470 * --install[:service_name]
471 * --uninstall[:service_name]
474 * They work as follows:
475 * --install will create a service for you and then terminate.
476 * By default the service name will be "privoxy" (without the quotes).
477 * However you can run multiple services if you wish, just by adding
478 * a colon and then a name (no spaces).
480 * --uninstall follows the exact same rules a --install.
482 * --service is used when the program is executed by the service
483 * control manager, and in normal circumstances would never be
484 * used as a command line argument.
486 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
487 * Windows service integration
489 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
490 * Incorporate Fabian Keil's patch work:
491 http://www.fabiankeil.de/sourcecode/privoxy/
493 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
494 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
495 * with what was really the latest development (the v_3_0_branch branch)
497 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
498 * Thanks to Jindrich Makovicka for a race condition fix for the log
499 * file. The race condition remains for non-pthread implementations.
500 * Reference patch #1175720.
502 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
503 * Add the ability to check jpeg images for invalid
504 * lengths of comment blocks. Defensive strategy
505 * against the exploit:
506 * Microsoft Security Bulletin MS04-028
507 * Buffer Overrun in JPEG Processing (GDI+) Could
508 * Allow Code Execution (833987)
509 * Enabled with +inspect-jpegs in actions files.
511 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
512 * - Fixed usage info for non-unix platforms
513 * - Fixed small cmdline parsing bug
515 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
516 * Diagnostics: Now preserve the returncode of pthread_create
517 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
519 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
520 * No longer ignore SIGCHLD. Fixes bug #769381
522 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
523 * Close jarfile on graceful exit, remove stray line
525 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
526 * Cosmetics: Killed a warning, a typo and an allocation left at exit
528 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
529 * No longer rely on non-POSIX.1 extensions of getcwd().
532 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
533 * Replaced setenv() by posix-compliant putenv()
534 * Thanks to Neil McCalden (nmcc AT users.sf.net).
536 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
537 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
539 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
540 * Clean-up and extension of improvements for forked mode:
541 * - Child's return code now consists of flags RC_FLAG_*
542 * - Reporting toggle to parent now properly #ifdef'ed
543 * - Children now report blocking to parent. This enables
544 * statistics in forked mode
546 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
547 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
548 * Children now report having been toggled through _exit(17), parents
549 * watch for that code and toggle themselves if found.
551 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
552 * Wrapping all *_r functions (the non-_r versions of them) with
553 * mutex semaphores for OSX. Hopefully this will take care of all
554 * of those pesky crash reports.
556 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
557 * Fixed two mostly harmless mem leaks
559 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
560 * Fix: Head of global clients list now initialized to NULL
562 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
563 * Added basic support for OPTIONS and TRACE HTTP methods:
564 * - New interceptor direct_response() added in chat().
565 * - sed() moved to earlier in the process, so that the
566 * Host: header is evaluated before actions and forwarding
569 * Revision 1.92 2002/05/08 16:00:46 oes
570 * Chat's buffer handling:
571 * - Fixed bug with unchecked out-of-mem conditions
572 * while reading client request & server headers
573 * - No longer predict if the buffer limit will be exceeded
574 * in the next read -- check add_to_iob's new
575 * return code. If buffer couldn't be extended
576 * (policy or out-of-mem) while
577 * - reading from client: abort
578 * - reading server headers: send error page
579 * - buffering server body for filter: flush,
580 * and if that fails: send error page
582 * Revision 1.91 2002/04/08 20:35:58 swa
585 * Revision 1.90 2002/04/02 14:57:28 oes
586 * Made sending wafers independent of FEATURE_COOKIE_JAR
588 * Revision 1.89 2002/03/31 17:18:59 jongfoster
589 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
591 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
592 * More compiler warning message maintenance
594 * Revision 1.87 2002/03/26 22:29:54 swa
595 * we have a new homepage!
597 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
598 * Workaround for closing the jarfile before load_config() comes around again
600 * Revision 1.85 2002/03/24 15:23:33 jongfoster
603 * Revision 1.84 2002/03/24 13:25:43 swa
604 * name change related issues
606 * Revision 1.83 2002/03/16 23:54:06 jongfoster
607 * Adding graceful termination feature, to help look for memory leaks.
608 * If you enable this (which, by design, has to be done by hand
609 * editing config.h) and then go to http://i.j.b/die, then the program
610 * will exit cleanly after the *next* request. It should free all the
611 * memory that was used.
613 * Revision 1.82 2002/03/13 00:27:05 jongfoster
616 * Revision 1.81 2002/03/12 01:42:50 oes
617 * Introduced modular filters
619 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
620 * OS/2 port maintenance:
621 * - Fixed EMX build - it had decayed a little
622 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
623 * substituted a memset for now.
625 * Revision 1.79 2002/03/09 20:03:52 jongfoster
626 * - Making various functions return int rather than size_t.
627 * (Undoing a recent change). Since size_t is unsigned on
628 * Windows, functions like read_socket that return -1 on
629 * error cannot return a size_t.
631 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
632 * crashes, and also frequently caused JB to jump to 100%
633 * CPU and stay there. (Because it thought it had just
634 * read ((unsigned)-1) == 4Gb of data...)
636 * - The signature of write_socket has changed, it now simply
637 * returns success=0/failure=nonzero.
639 * - Trying to get rid of a few warnings --with-debug on
640 * Windows, I've introduced a new type "jb_socket". This is
641 * used for the socket file descriptors. On Windows, this
642 * is SOCKET (a typedef for unsigned). Everywhere else, it's
643 * an int. The error value can't be -1 any more, so it's
644 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
645 * Windows it maps to the #define INVALID_SOCKET.)
647 * - The signature of bind_port has changed.
649 * Revision 1.78 2002/03/08 21:35:04 oes
650 * Added optional group supplement to --user option. Will now use default group of user if no group given
652 * Revision 1.77 2002/03/07 03:52:06 oes
653 * - Fixed compiler warnings etc
654 * - Improved handling of failed DNS lookups
656 * Revision 1.76 2002/03/06 22:54:35 jongfoster
657 * Automated function-comment nitpicking.
659 * Revision 1.75 2002/03/06 10:02:19 oes
660 * Fixed stupid bug when --user was not given
662 * Revision 1.74 2002/03/06 00:49:31 jongfoster
663 * Fixing warning on Windows
664 * Making #ifdefs that refer to the same variable consistently
665 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
667 * Revision 1.73 2002/03/05 23:57:30 hal9
668 * Stray character 's' on line 1618 was breaking build.
670 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
671 * - Re-enable OS/2 building after new parms were added
672 * - Fix false out of memory report when resolving CGI templates when no IP
673 * address is available of failed attempt (a la no such domain)
675 * Revision 1.71 2002/03/05 18:13:56 oes
676 * Added --user option
678 * Revision 1.70 2002/03/05 04:52:42 oes
679 * Deleted non-errlog debugging code
681 * Revision 1.69 2002/03/04 23:50:00 jongfoster
682 * Splitting off bind_port() call into bind_port_helper(), with
685 * Revision 1.68 2002/03/04 20:17:32 oes
688 * Revision 1.67 2002/03/04 18:18:57 oes
689 * - Removed _DEBUG mode
690 * - Cleand up cmdline parsing
691 * - Introduced --no-daemon, --pidfile options
692 * - Cleaned up signal handling:
693 * - Terminate cleanly on INT, TERM and ABRT
694 * - Schedule logfile for re-opening on HUP
695 * - Ignore CHLD and PIPE
696 * - Leave the rest with their default handlers
697 * - Uniform handler registration
698 * - Added usage() function
699 * - Played styleguide police
701 * Revision 1.66 2002/03/03 15:06:55 oes
702 * Re-enabled automatic config reloading
704 * Revision 1.65 2002/03/03 14:49:11 oes
705 * Fixed CLF logging: Now uses client's original HTTP request
707 * Revision 1.64 2002/03/03 09:18:03 joergs
708 * Made jumbjuster work on AmigaOS again.
710 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
711 * Clean up a little CRLF unpleasantness that suddenly appeared
713 * Revision 1.62 2002/02/20 23:17:23 jongfoster
714 * Detecting some out-of memory conditions and exiting with a log message.
716 * Revision 1.61 2002/01/17 21:01:52 jongfoster
717 * Moving all our URL and URL pattern parsing code to urlmatch.c.
719 * Revision 1.60 2001/12/30 14:07:32 steudten
720 * - Add signal handling (unix)
721 * - Add SIGHUP handler (unix)
722 * - Add creation of pidfile (unix)
723 * - Add action 'top' in rc file (RH)
724 * - Add entry 'SIGNALS' to manpage
725 * - Add exit message to logfile (unix)
727 * Revision 1.59 2001/12/13 14:07:18 oes
728 * Fixed Bug: 503 error page now sent OK
730 * Revision 1.58 2001/11/30 23:37:24 jongfoster
731 * Renaming the Win32 config file to config.txt - this is almost the
732 * same as the corresponding UNIX name "config"
734 * Revision 1.57 2001/11/16 00:47:43 jongfoster
735 * Changing the tty-disconnection code to use setsid().
737 * Revision 1.56 2001/11/13 20:20:54 jongfoster
738 * Tabs->spaces, fixing a bug with missing {} around an if()
740 * Revision 1.55 2001/11/13 20:14:53 jongfoster
741 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
743 * Revision 1.54 2001/11/07 00:03:14 steudten
744 * Give reliable return value if an error
745 * occurs not just 0 with new daemon mode.
747 * Revision 1.53 2001/11/05 21:41:43 steudten
748 * Add changes to be a real daemon just for unix os.
749 * (change cwd to /, detach from controlling tty, set
750 * process group and session leader to the own process.
752 * Add some fatal-error log message for failed malloc().
753 * Add '-d' if compiled with 'configure --with-debug' to
754 * enable debug output.
756 * Revision 1.52 2001/10/26 20:11:20 jongfoster
757 * Fixing type mismatch
759 * Revision 1.51 2001/10/26 17:38:28 oes
762 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
763 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
764 * threads to call select() simultaneously. So, it's time to do a real, live,
765 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
766 * (native). Both versions will work, but using __OS2__ offers multi-threading.
768 * Revision 1.49 2001/10/23 21:41:35 jongfoster
769 * Added call to initialize the (statically-allocated of course)
770 * "out of memory" CGI response.
772 * Revision 1.48 2001/10/10 19:56:46 jongfoster
773 * Moving some code that wasn't cookie-related out of an #ifdef
776 * Revision 1.47 2001/10/10 16:44:36 oes
777 * Added CONNECT destination port limitation check
779 * Revision 1.46 2001/10/08 15:17:41 oes
780 * Re-enabled SSL forwarding
782 * Revision 1.45 2001/10/07 15:42:11 oes
783 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
785 * Moved downgrading of the HTTP version from parse_http_request to
786 * chat(), since we can't decide if it is necessary before we have
787 * determined the actions for the URL. The HTTP command is now
788 * *always* re-built so the repairs need no longer be special-cased.
790 * filter_popups now gets a csp pointer so it can raise the new
791 * CSP_FLAG_MODIFIED flag.
795 * Added configurable size limit for the IOB. If the IOB grows so
796 * large that the next read would exceed the limit, the header
797 * is generated, and the header & unfiltered buffer are flushed
798 * to the client. Chat then continues in non-buffering,
799 * non-filtering body mode.
801 * Revision 1.44 2001/10/02 18:13:57 oes
804 * Revision 1.43 2001/10/02 15:32:13 oes
805 * Moved generation of hdr
807 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
808 * Cleaning up 2 compiler warnings on OS/2.
810 * Revision 1.41 2001/09/16 17:05:14 jongfoster
811 * Removing unused #include showarg.h
813 * Revision 1.40 2001/09/16 15:41:45 jongfoster
814 * Fixing signed/unsigned comparison warning.
816 * Revision 1.39 2001/09/16 13:21:27 jongfoster
817 * Changes to use new list functions.
819 * Revision 1.38 2001/09/16 13:01:46 jongfoster
820 * Removing redundant function call that zeroed zalloc()'d memory.
822 * Revision 1.37 2001/09/10 11:12:24 oes
823 * Deleted unused variable
825 * Revision 1.36 2001/09/10 10:56:15 oes
826 * Silenced compiler warnings
828 * Revision 1.35 2001/07/31 14:44:22 oes
829 * Deleted unused size parameter from filter_popups()
831 * Revision 1.34 2001/07/30 22:08:36 jongfoster
832 * Tidying up #defines:
833 * - All feature #defines are now of the form FEATURE_xxx
834 * - Permanently turned off WIN_GUI_EDIT
835 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
837 * Revision 1.33 2001/07/29 19:32:00 jongfoster
838 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
840 * Revision 1.32 2001/07/29 18:47:05 jongfoster
841 * Adding missing #include "loadcfg.h"
843 * Revision 1.31 2001/07/29 12:17:48 oes
844 * Applied pthread fix by Paul Lieverse
846 * Revision 1.30 2001/07/25 22:57:13 jongfoster
847 * __BEOS__ no longer overrides FEATURE_PTHREAD.
848 * This is because FEATURE_PTHREAD will soon be widely used, so I
849 * want to keep it simple.
851 * Revision 1.29 2001/07/24 12:47:06 oes
852 * Applied BeOS support update by Eugenia
854 * Revision 1.28 2001/07/23 13:26:12 oes
855 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
857 * Revision 1.27 2001/07/19 19:09:47 haroon
858 * - Added code to take care of the situation where while processing the first
859 * server response (which includes the server header), after finding the end
860 * of the headers we were not looking past the end of the headers for
861 * content modification. I enabled it for filter_popups.
862 * Someone else should look to see if other similar operations should be
863 * done to the discarded portion of the buffer.
865 * Note 2001/07/20: No, the other content modification mechanisms will process
866 * the whole iob later anyway. --oes
868 * Revision 1.26 2001/07/18 12:31:36 oes
871 * Revision 1.25 2001/07/15 19:43:49 jongfoster
872 * Supports POSIX threads.
873 * Also removed some unused #includes.
875 * Revision 1.24 2001/07/13 14:00:40 oes
876 * - Generic content modification scheme:
877 * Each feature has its own applicability flag that is set
878 * from csp->action->flags.
879 * Replaced the "filtering" int flag , by a function pointer
880 * "content_filter" to the function that will do the content
881 * modification. If it is != NULL, the document will be buffered
882 * and processed through *content_filter, which must set
883 * csp->content_length and return a modified copy of the body
884 * or return NULL (on failiure).
885 * - Changed csp->is_text to the more generic bitmap csp->content_type
886 * which can currently take the valued CT_TEXT or CT_GIF
888 * - Removed all #ifdef PCRS
890 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
891 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
892 * the `block_acl' call. This prevents a compilation error when the user
893 * does not wish to use the "ACL" feature.
895 * Revision 1.22 2001/06/29 21:45:41 oes
896 * Indentation, CRLF->LF, Tab-> Space
898 * Revision 1.21 2001/06/29 13:29:36 oes
899 * - Cleaned up, improved comments
900 * - Unified all possible interceptors (CGI,
901 * block, trust, fast_redirect) in one
902 * place, with one (CGI) answer generation
903 * mechansim. Much clearer now.
904 * - Removed the GIF image generation, which
905 * is now done in filters.c:block_url()
906 * - Made error conditions like domain lookup
907 * failiure or (various) problems while talking
908 * to the server use cgi.c:error_response()
909 * instead of generating HTML/HTTP in chat() (yuck!)
910 * - Removed logentry from cancelled commit
912 * Revision 1.20 2001/06/09 10:55:28 jongfoster
913 * Changing BUFSIZ ==> BUFFER_SIZE
915 * Revision 1.19 2001/06/07 23:12:52 jongfoster
916 * Replacing function pointer in struct gateway with a directly
917 * called function forwarded_connect().
918 * Replacing struct gateway with struct forward_spec
920 * Revision 1.18 2001/06/03 19:12:16 oes
921 * introduced new cgi handling
923 * Revision 1.17 2001/06/01 20:07:23 jongfoster
924 * Now uses action +image-blocker{} rather than config->tinygif
926 * Revision 1.16 2001/06/01 18:49:17 jongfoster
927 * Replaced "list_share" with "list" - the tiny memory gain was not
928 * worth the extra complexity.
930 * Revision 1.15 2001/05/31 21:24:47 jongfoster
931 * Changed "permission" to "action" throughout.
932 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
933 * Moved vanilla wafer check into chat(), since we must now
934 * decide whether or not to add it based on the URL.
936 * Revision 1.14 2001/05/29 20:14:01 joergs
937 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
940 * Revision 1.13 2001/05/29 09:50:24 jongfoster
941 * Unified blocklist/imagelist/permissionslist.
942 * File format is still under discussion, but the internal changes
945 * Also modified interceptor behaviour:
946 * - We now intercept all URLs beginning with one of the following
947 * prefixes (and *only* these prefixes):
949 * * http://ijbswa.sf.net/config/
950 * * http://ijbswa.sourceforge.net/config/
951 * - New interceptors "home page" - go to http://i.j.b/ to see it.
952 * - Internal changes so that intercepted and fast redirect pages
953 * are not replaced with an image.
954 * - Interceptors now have the option to send a binary page direct
955 * to the client. (i.e. ijb-send-banner uses this)
956 * - Implemented show-url-info interceptor. (Which is why I needed
957 * the above interceptors changes - a typical URL is
958 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
959 * The previous mechanism would not have intercepted that, and
960 * if it had been intercepted then it then it would have replaced
963 * Revision 1.12 2001/05/27 22:17:04 oes
965 * - re_process_buffer no longer writes the modified buffer
966 * to the client, which was very ugly. It now returns the
967 * buffer, which it is then written by chat.
969 * - content_length now adjusts the Content-Length: header
970 * for modified documents rather than crunch()ing it.
971 * (Length info in csp->content_length, which is 0 for
972 * unmodified documents)
974 * - For this to work, sed() is called twice when filtering.
976 * Revision 1.11 2001/05/26 17:27:53 jongfoster
977 * Added support for CLF and fixed LOG_LEVEL_LOG.
978 * Also did CRLF->LF fix of my previous patch.
980 * Revision 1.10 2001/05/26 15:26:15 jongfoster
981 * ACL feature now provides more security by immediately dropping
982 * connections from untrusted hosts.
984 * Revision 1.9 2001/05/26 00:28:36 jongfoster
985 * Automatic reloading of config file.
986 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
987 * Most of the global variables have been moved to a new
988 * struct configuration_spec, accessed through csp->config->globalname
989 * Most of the globals remaining are used by the Win32 GUI.
991 * Revision 1.8 2001/05/25 22:43:18 jongfoster
992 * Fixing minor memory leak and buffer overflow.
994 * Revision 1.7 2001/05/25 22:34:30 jongfoster
997 * Revision 1.6 2001/05/23 00:13:58 joergs
998 * AmigaOS support fixed.
1000 * Revision 1.5 2001/05/22 18:46:04 oes
1002 * - Enabled filtering banners by size rather than URL
1003 * by adding patterns that replace all standard banner
1004 * sizes with the "Junkbuster" gif to the re_filterfile
1006 * - Enabled filtering WebBugs by providing a pattern
1007 * which kills all 1x1 images
1009 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
1010 * which is selected by the (nonstandard and therefore
1011 * capital) letter 'U' in the option string.
1012 * It causes the quantifiers to be ungreedy by default.
1013 * Appending a ? turns back to greedy (!).
1015 * - Added a new interceptor ijb-send-banner, which
1016 * sends back the "Junkbuster" gif. Without imagelist or
1017 * MSIE detection support, or if tinygif = 1, or the
1018 * URL isn't recognized as an imageurl, a lame HTML
1019 * explanation is sent instead.
1021 * - Added new feature, which permits blocking remote
1022 * script redirects and firing back a local redirect
1024 * The feature is conditionally compiled, i.e. it
1025 * can be disabled with --disable-fast-redirects,
1026 * plus it must be activated by a "fast-redirects"
1027 * line in the config file, has its own log level
1028 * and of course wants to be displayed by show-proxy-args
1029 * Note: Boy, all the #ifdefs in 1001 locations and
1030 * all the fumbling with configure.in and acconfig.h
1031 * were *way* more work than the feature itself :-(
1033 * - Because a generic redirect template was needed for
1034 * this, tinygif = 3 now uses the same.
1036 * - Moved GIFs, and other static HTTP response templates
1039 * - Some minor fixes
1041 * - Removed some >400 CRs again (Jon, you really worked
1044 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1045 * Made failure to bind() a fatal error.
1047 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1048 * Version 2.9.4 checkin.
1049 * - Merged popupfile and cookiefile, and added control over PCRS
1050 * filtering, in new "permissionsfile".
1051 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1052 * file error you now get a message box (in the Win32 GUI) rather
1053 * than the program exiting with no explanation.
1054 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1056 * - Removed tabs from "config"
1057 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1058 * - Bumped up version number.
1060 * Revision 1.2 2001/05/17 22:34:44 oes
1061 * - Added hint on GIF char array generation to jcc.c
1062 * - Cleaned CRLF's from the sources and related files
1063 * - Repaired logging for REF and FRC
1065 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1066 * Initial import of version 2.9.3 source tree
1069 *********************************************************************/
1075 #include <sys/types.h>
1084 # ifndef FEATURE_PTHREAD
1088 # include <windows.h>
1089 # include <process.h>
1090 # endif /* ndef FEATURE_PTHREAD */
1093 # ifndef _WIN_CONSOLE
1094 # include "w32log.h"
1095 # endif /* ndef _WIN_CONSOLE */
1096 # include "w32svrapi.h"
1098 #else /* ifndef _WIN32 */
1100 # if !defined (__OS2__)
1101 # include <unistd.h>
1102 # include <sys/wait.h>
1103 # endif /* ndef __OS2__ */
1104 # include <sys/time.h>
1105 # include <sys/stat.h>
1106 # include <sys/ioctl.h>
1109 #include <sys/termios.h>
1117 # include <signal.h>
1120 # include <socket.h> /* BeOS has select() for sockets only. */
1121 # include <OS.h> /* declarations for threads and stuff. */
1124 # if defined(__EMX__) || defined(__OS2__)
1125 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1130 #define bzero(B,N) memset(B,0x00,n)
1134 # include <select.h>
1139 #include "project.h"
1142 #include "filters.h"
1143 #include "loaders.h"
1144 #include "parsers.h"
1145 #include "miscutil.h"
1147 #include "jbsockets.h"
1148 #include "gateway.h"
1149 #include "actions.h"
1151 #include "loadcfg.h"
1152 #include "urlmatch.h"
1154 const char jcc_h_rcs[] = JCC_H_VERSION;
1155 const char project_h_rcs[] = PROJECT_H_VERSION;
1158 struct client_state clients[1];
1159 struct file_list files[1];
1161 #ifdef FEATURE_STATISTICS
1162 int urls_read = 0; /* total nr of urls read inc rejected */
1163 int urls_rejected = 0; /* total nr of urls rejected */
1164 #endif /* def FEATURE_STATISTICS */
1166 #ifdef FEATURE_GRACEFUL_TERMINATION
1167 int g_terminate = 0;
1170 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1171 static void sig_handler(int the_signal);
1173 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1174 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1175 static jb_err get_server_headers(struct client_state *csp);
1176 static const char *crunch_reason(const struct http_response *rsp);
1177 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1179 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1181 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1182 static jb_err change_request_destination(struct client_state *csp);
1183 static void chat(struct client_state *csp);
1184 static void serve(struct client_state *csp);
1185 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1186 static void usage(const char *myname);
1188 static void initialize_mutexes(void);
1189 static jb_socket bind_port_helper(struct configuration_spec *config);
1190 static void listen_loop(void);
1193 void serve(struct client_state *csp);
1194 #else /* ifndef AMIGA */
1195 static void serve(struct client_state *csp);
1196 #endif /* def AMIGA */
1199 static int32 server_thread(void *data);
1200 #endif /* def __BEOS__ */
1203 #define sleep(N) Sleep(((N) * 1000))
1207 #define sleep(N) DosSleep(((N) * 100))
1210 #ifdef MUTEX_LOCKS_AVAILABLE
1212 * XXX: Does the locking stuff really belong in this file?
1214 privoxy_mutex_t log_mutex;
1215 privoxy_mutex_t log_init_mutex;
1216 privoxy_mutex_t connection_reuse_mutex;
1218 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1219 privoxy_mutex_t resolver_mutex;
1220 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1222 #ifndef HAVE_GMTIME_R
1223 privoxy_mutex_t gmtime_mutex;
1224 #endif /* ndef HAVE_GMTIME_R */
1226 #ifndef HAVE_LOCALTIME_R
1227 privoxy_mutex_t localtime_mutex;
1228 #endif /* ndef HAVE_GMTIME_R */
1231 privoxy_mutex_t rand_mutex;
1232 #endif /* ndef HAVE_RANDOM */
1234 #endif /* def MUTEX_LOCKS_AVAILABLE */
1237 const char *basedir = NULL;
1238 const char *pidfile = NULL;
1239 static int received_hup_signal = 0;
1240 #endif /* defined unix */
1242 /* HTTP snipplets. */
1243 static const char CSUCCEED[] =
1244 "HTTP/1.0 200 Connection established\r\n"
1245 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1247 static const char CHEADER[] =
1248 "HTTP/1.0 400 Invalid header received from client\r\n"
1249 "Proxy-Agent: Privoxy " VERSION "\r\n"
1250 "Content-Type: text/plain\r\n"
1251 "Connection: close\r\n\r\n"
1252 "Invalid header received from client.\r\n";
1254 static const char FTP_RESPONSE[] =
1255 "HTTP/1.0 400 Invalid request received from client\r\n"
1256 "Content-Type: text/plain\r\n"
1257 "Connection: close\r\n\r\n"
1258 "Invalid request. Privoxy doesn't support FTP.\r\n";
1260 static const char GOPHER_RESPONSE[] =
1261 "HTTP/1.0 400 Invalid request received from client\r\n"
1262 "Content-Type: text/plain\r\n"
1263 "Connection: close\r\n\r\n"
1264 "Invalid request. Privoxy doesn't support gopher.\r\n";
1266 /* XXX: should be a template */
1267 static const char MISSING_DESTINATION_RESPONSE[] =
1268 "HTTP/1.0 400 Bad request received from client\r\n"
1269 "Proxy-Agent: Privoxy " VERSION "\r\n"
1270 "Content-Type: text/plain\r\n"
1271 "Connection: close\r\n\r\n"
1272 "Bad request. Privoxy was unable to extract the destination.\r\n";
1274 /* XXX: should be a template */
1275 static const char NO_SERVER_DATA_RESPONSE[] =
1276 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1277 "Proxy-Agent: Privoxy " VERSION "\r\n"
1278 "Content-Type: text/plain\r\n"
1279 "Connection: close\r\n\r\n"
1280 "Empty server or forwarder response.\r\n"
1281 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1283 /* XXX: should be a template */
1284 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1285 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1286 "Proxy-Agent: Privoxy " VERSION "\r\n"
1287 "Content-Type: text/plain\r\n"
1288 "Connection: close\r\n\r\n"
1289 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1292 /* XXX: should be a template */
1293 static const char NULL_BYTE_RESPONSE[] =
1294 "HTTP/1.0 400 Bad request received from client\r\n"
1295 "Proxy-Agent: Privoxy " VERSION "\r\n"
1296 "Content-Type: text/plain\r\n"
1297 "Connection: close\r\n\r\n"
1298 "Bad request. Null byte(s) before end of request.\r\n";
1301 /* XXX: should be a template */
1302 static const char MESSED_UP_REQUEST_RESPONSE[] =
1303 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1304 "Proxy-Agent: Privoxy " VERSION "\r\n"
1305 "Content-Type: text/plain\r\n"
1306 "Connection: close\r\n\r\n"
1307 "Bad request. Messed up with header filters.\r\n";
1309 /* A function to crunch a response */
1310 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1312 /* Crunch function flags */
1313 #define CF_NO_FLAGS 0
1314 /* Cruncher applies to forced requests as well */
1315 #define CF_IGNORE_FORCE 1
1316 /* Crunched requests are counted for the block statistics */
1317 #define CF_COUNT_AS_REJECT 2
1319 /* A crunch function and its flags */
1322 const crunch_func_ptr cruncher;
1326 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1328 /* Complete list of cruncher functions */
1329 static const struct cruncher crunchers_all[] = {
1330 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1331 { block_url, CF_COUNT_AS_REJECT },
1332 #ifdef FEATURE_TRUST
1333 { trust_url, CF_COUNT_AS_REJECT },
1334 #endif /* def FEATURE_TRUST */
1335 { redirect_url, CF_NO_FLAGS },
1336 { dispatch_cgi, CF_IGNORE_FORCE},
1340 /* Light version, used after tags are applied */
1341 static const struct cruncher crunchers_light[] = {
1342 { block_url, CF_COUNT_AS_REJECT },
1343 { redirect_url, CF_NO_FLAGS },
1349 * XXX: Don't we really mean
1355 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1356 /*********************************************************************
1358 * Function : sig_handler
1360 * Description : Signal handler for different signals.
1361 * Exit gracefully on TERM and INT
1362 * or set a flag that will cause the errlog
1363 * to be reopened by the main thread on HUP.
1366 * 1 : the_signal = the signal cause this function to call
1370 *********************************************************************/
1371 static void sig_handler(int the_signal)
1377 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1389 received_hup_signal = 1;
1395 * We shouldn't be here, unless we catch signals
1396 * in main() that we can't handle here!
1398 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1406 /*********************************************************************
1408 * Function : client_protocol_is_unsupported
1410 * Description : Checks if the client used a known unsupported
1411 * protocol and deals with it by sending an error
1415 * 1 : csp = Current client state (buffers, headers, etc...)
1416 * 2 : req = the first request line send by the client
1418 * Returns : TRUE if an error response has been generated, or
1419 * FALSE if the request doesn't look invalid.
1421 *********************************************************************/
1422 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1425 * If it's a FTP or gopher request, we don't support it.
1427 * These checks are better than nothing, but they might
1428 * not work in all configurations and some clients might
1429 * have problems digesting the answer.
1431 * They should, however, never cause more problems than
1432 * Privoxy's old behaviour (returning the misleading HTML
1435 * "Could not resolve http://(ftp|gopher)://example.org").
1437 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1439 const char *response = NULL;
1440 const char *protocol = NULL;
1442 if (!strncmpic(req, "GET ftp://", 10))
1444 response = FTP_RESPONSE;
1449 response = GOPHER_RESPONSE;
1450 protocol = "GOPHER";
1452 log_error(LOG_LEVEL_ERROR,
1453 "%s tried to use Privoxy as %s proxy: %s",
1454 csp->ip_addr_str, protocol, req);
1455 log_error(LOG_LEVEL_CLF,
1456 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1458 write_socket(csp->cfd, response, strlen(response));
1467 /*********************************************************************
1469 * Function : get_request_destination_elsewhere
1471 * Description : If the client's request was redirected into
1472 * Privoxy without the client's knowledge,
1473 * the request line lacks the destination host.
1475 * This function tries to get it elsewhere,
1476 * provided accept-intercepted-requests is enabled.
1478 * "Elsewhere" currently only means "Host: header",
1479 * but in the future we may ask the redirecting
1480 * packet filter to look the destination up.
1482 * If the destination stays unknown, an error
1483 * response is send to the client and headers
1484 * are freed so that chat() can return directly.
1487 * 1 : csp = Current client state (buffers, headers, etc...)
1488 * 2 : headers = a header list
1490 * Returns : JB_ERR_OK if the destination is now known, or
1491 * JB_ERR_PARSE if it isn't.
1493 *********************************************************************/
1494 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1498 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1500 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1501 " Privoxy isn't configured to accept intercepted requests.",
1502 csp->ip_addr_str, csp->http->cmd);
1503 /* XXX: Use correct size */
1504 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1505 csp->ip_addr_str, csp->http->cmd);
1507 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1508 destroy_list(headers);
1510 return JB_ERR_PARSE;
1512 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1514 /* Split the domain we just got for pattern matching */
1515 init_domain_components(csp->http);
1521 /* We can't work without destination. Go spread the news.*/
1523 req = list_to_text(headers);
1525 /* XXX: Use correct size */
1526 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1527 csp->ip_addr_str, csp->http->cmd);
1528 log_error(LOG_LEVEL_ERROR,
1529 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1530 csp->ip_addr_str, csp->http->cmd, req);
1533 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1534 destroy_list(headers);
1536 return JB_ERR_PARSE;
1539 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1540 * to get the destination IP address, use it as host directly
1541 * or do a reverse DNS lookup first.
1546 /*********************************************************************
1548 * Function : get_server_headers
1550 * Description : Parses server headers in iob and fills them
1551 * into csp->headers so that they can later be
1555 * 1 : csp = Current client state (buffers, headers, etc...)
1557 * Returns : JB_ERR_OK if everything went fine, or
1558 * JB_ERR_PARSE if the headers were incomplete.
1560 *********************************************************************/
1561 static jb_err get_server_headers(struct client_state *csp)
1563 int continue_hack_in_da_house = 0;
1566 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1571 * continue hack in da house. Ignore the ending of
1572 * this head and continue enlisting header lines.
1573 * The reason is described below.
1575 enlist(csp->headers, "");
1576 continue_hack_in_da_house = 0;
1579 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1582 * It's a bodyless continue response, don't
1583 * stop header parsing after reaching its end.
1585 * As a result Privoxy will concatenate the
1586 * next response's head and parse and deliver
1587 * the headers as if they belonged to one request.
1589 * The client will separate them because of the
1590 * empty line between them.
1592 * XXX: What we're doing here is clearly against
1593 * the intended purpose of the continue header,
1594 * and under some conditions (HTTP/1.0 client request)
1595 * it's a standard violation.
1597 * Anyway, "sort of against the spec" is preferable
1598 * to "always getting confused by Continue responses"
1599 * (Privoxy's behaviour before this hack was added)
1601 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1602 continue_hack_in_da_house = 1;
1604 else if (*header == '\0')
1607 * If the header is empty, but the Continue hack
1608 * isn't active, we can assume that we reached the
1609 * end of the buffer before we hit the end of the
1612 * Inform the caller an let it decide how to handle it.
1614 return JB_ERR_PARSE;
1617 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1620 * XXX: Should we quit the request and return a
1621 * out of memory error page instead?
1623 log_error(LOG_LEVEL_ERROR,
1624 "Out of memory while enlisting server headers. %s lost.",
1634 /*********************************************************************
1636 * Function : crunch_reason
1638 * Description : Translates the crunch reason code into a string.
1641 * 1 : rsp = a http_response
1643 * Returns : A string with the crunch reason or an error description.
1645 *********************************************************************/
1646 static const char *crunch_reason(const struct http_response *rsp)
1648 char * reason = NULL;
1650 assert(rsp != NULL);
1653 return "Internal error while searching for crunch reason";
1656 switch (rsp->reason)
1658 case RSP_REASON_UNSUPPORTED:
1659 reason = "Unsupported HTTP feature";
1661 case RSP_REASON_BLOCKED:
1664 case RSP_REASON_UNTRUSTED:
1665 reason = "Untrusted";
1667 case RSP_REASON_REDIRECTED:
1668 reason = "Redirected";
1670 case RSP_REASON_CGI_CALL:
1671 reason = "CGI Call";
1673 case RSP_REASON_NO_SUCH_DOMAIN:
1674 reason = "DNS failure";
1676 case RSP_REASON_FORWARDING_FAILED:
1677 reason = "Forwarding failed";
1679 case RSP_REASON_CONNECT_FAILED:
1680 reason = "Connection failure";
1682 case RSP_REASON_OUT_OF_MEMORY:
1683 reason = "Out of memory (may mask other reasons)";
1686 reason = "No reason recorded";
1694 /*********************************************************************
1696 * Function : send_crunch_response
1698 * Description : Delivers already prepared response for
1699 * intercepted requests, logs the interception
1700 * and frees the response.
1703 * 1 : csp = Current client state (buffers, headers, etc...)
1704 * 1 : rsp = Fully prepared response. Will be freed on exit.
1706 * Returns : Nothing.
1708 *********************************************************************/
1709 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1711 const struct http_request *http = csp->http;
1712 char status_code[4];
1714 assert(rsp != NULL);
1715 assert(rsp->head != NULL);
1720 * Not supposed to happen. If it does
1721 * anyway, treat it as an unknown error.
1723 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1724 /* return code doesn't matter */
1729 /* If rsp is still NULL, we have serious internal problems. */
1730 log_error(LOG_LEVEL_FATAL,
1731 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1735 * Extract the status code from the actual head
1736 * that was send to the client. It is the only
1737 * way to get it right for all requests, including
1738 * the fixed ones for out-of-memory problems.
1740 * A head starts like this: 'HTTP/1.1 200...'
1744 status_code[0] = rsp->head[9];
1745 status_code[1] = rsp->head[10];
1746 status_code[2] = rsp->head[11];
1747 status_code[3] = '\0';
1749 /* Write the answer to the client */
1750 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1751 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1753 /* There is nothing we can do about it. */
1754 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1757 /* Log that the request was crunched and why. */
1758 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1759 http->hostport, http->path, crunch_reason(rsp));
1760 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1761 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1763 /* Clean up and return */
1764 if (cgi_error_memory() != rsp)
1766 free_http_response(rsp);
1773 /*********************************************************************
1775 * Function : request_contains_null_bytes
1777 * Description : Checks for NULL bytes in the request and sends
1778 * an error message to the client if any were found.
1780 * XXX: currently not used, see comment in chat().
1783 * 1 : csp = Current client state (buffers, headers, etc...)
1784 * 2 : buf = Data from the client's request to check.
1785 * 3 : len = The data length.
1787 * Returns : TRUE if the request contained one or more NULL bytes, or
1790 *********************************************************************/
1791 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1793 size_t c_len; /* Request lenght when treated as C string */
1795 c_len = strlen(buf);
1800 * Null byte(s) found. Log the request,
1801 * return an error response and hang up.
1803 size_t tmp_len = c_len;
1808 * Replace NULL byte(s) with '°' characters
1809 * so the request can be logged as string.
1810 * XXX: Is there a better replacement character?
1813 tmp_len += strlen(buf+tmp_len);
1814 } while (tmp_len < len);
1816 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1817 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1818 log_error(LOG_LEVEL_HEADER,
1819 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1821 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1823 /* XXX: Log correct size */
1824 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1834 /*********************************************************************
1836 * Function : crunch_response_triggered
1838 * Description : Checks if the request has to be crunched,
1839 * and delivers the crunch response if necessary.
1842 * 1 : csp = Current client state (buffers, headers, etc...)
1843 * 2 : crunchers = list of cruncher functions to run
1845 * Returns : TRUE if the request was answered with a crunch response
1848 *********************************************************************/
1849 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1851 struct http_response *rsp = NULL;
1852 const struct cruncher *c;
1855 * If CGI request crunching is disabled,
1856 * check the CGI dispatcher out of order to
1857 * prevent unintentional blocks or redirects.
1859 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1860 && (NULL != (rsp = dispatch_cgi(csp))))
1862 /* Deliver, log and free the interception response. */
1863 send_crunch_response(csp, rsp);
1867 for (c = crunchers; c->cruncher != NULL; c++)
1870 * Check the cruncher if either Privoxy is toggled
1871 * on and the request isn't forced, or if the cruncher
1872 * applies to forced requests as well.
1874 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1875 !(csp->flags & CSP_FLAG_FORCED)) ||
1876 (c->flags & CF_IGNORE_FORCE))
1878 rsp = c->cruncher(csp);
1881 /* Deliver, log and free the interception response. */
1882 send_crunch_response(csp, rsp);
1883 #ifdef FEATURE_STATISTICS
1884 if (c->flags & CF_COUNT_AS_REJECT)
1886 csp->flags |= CSP_FLAG_REJECTED;
1888 #endif /* def FEATURE_STATISTICS */
1899 /*********************************************************************
1901 * Function : build_request_line
1903 * Description : Builds the HTTP request line.
1905 * If a HTTP forwarder is used it expects the whole URL,
1906 * web servers only get the path.
1909 * 1 : csp = Current client state (buffers, headers, etc...)
1910 * 2 : fwd = The forwarding spec used for the request
1911 * XXX: Should use http->fwd instead.
1912 * 3 : request_line = The old request line which will be replaced.
1914 * Returns : Nothing. Terminates in case of memory problems.
1916 *********************************************************************/
1917 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1919 struct http_request *http = csp->http;
1921 assert(http->ssl == 0);
1924 * Downgrade http version from 1.1 to 1.0
1925 * if +downgrade action applies.
1927 if ( (csp->action->flags & ACTION_DOWNGRADE)
1928 && (!strcmpic(http->ver, "HTTP/1.1")))
1931 http->ver = strdup("HTTP/1.0");
1933 if (http->ver == NULL)
1935 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1940 * Rebuild the request line.
1942 freez(*request_line);
1943 *request_line = strdup(http->gpc);
1944 string_append(request_line, " ");
1946 if (fwd->forward_host)
1948 string_append(request_line, http->url);
1952 string_append(request_line, http->path);
1954 string_append(request_line, " ");
1955 string_append(request_line, http->ver);
1957 if (*request_line == NULL)
1959 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1961 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1965 /*********************************************************************
1967 * Function : change_request_destination
1969 * Description : Parse a (rewritten) request line and regenerate
1970 * the http request data.
1973 * 1 : csp = Current client state (buffers, headers, etc...)
1975 * Returns : Forwards the parse_http_request() return code.
1976 * Terminates in case of memory problems.
1978 *********************************************************************/
1979 static jb_err change_request_destination(struct client_state *csp)
1981 struct http_request *http = csp->http;
1984 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1985 free_http_request(http);
1986 err = parse_http_request(csp->headers->first->str, http, csp);
1987 if (JB_ERR_OK != err)
1989 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1990 jb_err_to_string(err));
1994 /* XXX: ocmd is a misleading name */
1995 http->ocmd = strdup(http->cmd);
1996 if (http->ocmd == NULL)
1998 log_error(LOG_LEVEL_FATAL,
1999 "Out of memory copying rewritten HTTP request line");
2007 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2008 /*********************************************************************
2010 * Function : server_response_is_complete
2012 * Description : Determines whether we should stop reading
2013 * from the server socket.
2016 * 1 : csp = Current client state (buffers, headers, etc...)
2017 * 2 : content_length = Length of content received so far.
2019 * Returns : TRUE if the response is complete,
2022 *********************************************************************/
2023 static int server_response_is_complete(struct client_state *csp, size_t content_length)
2025 int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2027 if (!strcmpic(csp->http->gpc, "HEAD"))
2030 * "HEAD" implies no body, we are thus expecting
2031 * no content. XXX: incomplete "list" of methods?
2033 log_error(LOG_LEVEL_INFO, "Method %s implies no body.", csp->http->gpc);
2034 csp->expected_content_length = 0;
2035 content_length_known = TRUE;
2038 if (csp->http->status == 304)
2041 * Expect no body. XXX: incomplete "list" of status codes?
2043 log_error(LOG_LEVEL_INFO, "Status code %d implies no body.", csp->http->status);
2044 csp->expected_content_length = 0;
2045 content_length_known = TRUE;
2048 return (content_length_known && ((0 == csp->expected_content_length)
2049 || (csp->expected_content_length <= content_length)));
2051 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2054 /*********************************************************************
2058 * Description : Once a connection to the client has been accepted,
2059 * this function is called (via serve()) to handle the
2060 * main business of the communication. When this
2061 * function returns, the caller must close the client
2064 * FIXME: chat is nearly thousand lines long.
2068 * 1 : csp = Current client state (buffers, headers, etc...)
2070 * Returns : Nothing.
2072 *********************************************************************/
2073 static void chat(struct client_state *csp)
2075 char buf[BUFFER_SIZE];
2083 int ms_iis5_hack = 0;
2084 size_t byte_count = 0;
2085 int forwarded_connect_retries = 0;
2086 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2087 const struct forward_spec * fwd;
2088 struct http_request *http;
2089 int len; /* for buffer sizes (and negative error codes) */
2092 /* Function that does the content filtering for the current request */
2093 filter_function_ptr content_filter = NULL;
2095 /* Skeleton for HTTP response, if we should intercept the request */
2096 struct http_response *rsp;
2098 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2099 struct list header_list;
2100 struct list *headers = &header_list;
2104 memset(buf, 0, sizeof(buf));
2107 * Read the client's request. Note that since we're not using select() we
2108 * could get blocked here if a client connected, then didn't say anything!
2113 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2115 if (len <= 0) break; /* error! */
2118 * If there is no memory left for buffering the
2119 * request, there is nothing we can do but hang up
2121 if (add_to_iob(csp, buf, len))
2126 req = get_header(csp->iob);
2128 } while ((NULL != req) && ('\0' == *req));
2130 if ((NULL != req) && ('\0' != *req))
2132 /* Request received. Validate and parse it. */
2136 * XXX: Temporary disabled to prevent problems
2137 * with POST requests whose bodies are allowed to
2138 * contain NULL bytes. BR#1730105.
2140 * The main purpose of this check is to properly
2141 * log stuff like BitTorrent traffic and other junk
2142 * that hits public proxies. It's not required for
2143 * Privoxy to functions as those requests are discarded
2146 * It probably should be rewritten to only check
2147 * the head of the request. Another option would
2148 * be to let all POST requests pass, although that
2149 * may not be good enough.
2151 if (request_contains_null_bytes(csp, buf, len))
2153 /* NULL bytes found and dealt with, just hang up. */
2158 /* Does the request line look invalid? */
2159 if (client_protocol_is_unsupported(csp, req))
2162 * Yes. The request has already been
2163 * answered with a error response, the buffers
2164 * were freed and we're done with chatting.
2169 #ifdef FEATURE_FORCE_LOAD
2171 * If this request contains the FORCE_PREFIX and blocks
2172 * aren't enforced, get rid of it and set the force flag.
2174 if (strstr(req, FORCE_PREFIX))
2176 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2178 log_error(LOG_LEVEL_FORCE,
2179 "Ignored force prefix in request: \"%s\".", req);
2183 strclean(req, FORCE_PREFIX);
2184 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2185 csp->flags |= CSP_FLAG_FORCED;
2189 #endif /* def FEATURE_FORCE_LOAD */
2190 err = parse_http_request(req, http, csp);
2191 if (JB_ERR_OK != err)
2193 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2199 if (http->cmd == NULL)
2201 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2202 /* XXX: Use correct size */
2203 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2204 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2206 free_http_request(http);
2210 /* grab the rest of the client's headers */
2214 p = get_header(csp->iob);
2218 /* There are no additional headers to read. */
2225 * We didn't receive a complete header
2226 * line yet, get the rest of it.
2228 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2231 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2232 destroy_list(headers);
2236 if (add_to_iob(csp, buf, len))
2239 * If there is no memory left for buffering the
2240 * request, there is nothing we can do but hang up
2242 destroy_list(headers);
2249 * We were able to read a complete
2250 * header and can finaly enlist it.
2257 if (http->host == NULL)
2260 * If we still don't know the request destination,
2261 * the request is invalid or the client uses
2262 * Privoxy without its knowledge.
2264 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2267 * Our attempts to get the request destination
2268 * elsewhere failed or Privoxy is configured
2269 * to only accept proxy requests.
2271 * An error response has already been send
2272 * and we're done here.
2279 * Determine the actions for this URL
2281 #ifdef FEATURE_TOGGLE
2282 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2284 /* Most compatible set of actions (i.e. none) */
2285 init_current_action(csp->action);
2288 #endif /* ndef FEATURE_TOGGLE */
2290 get_url_actions(csp, http);
2294 * Save a copy of the original request for logging
2296 http->ocmd = strdup(http->cmd);
2298 if (http->ocmd == NULL)
2300 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2303 enlist(csp->headers, http->cmd);
2305 /* Append the previously read headers */
2306 list_append_list_unique(csp->headers, headers);
2307 destroy_list(headers);
2309 err = sed(csp, FILTER_CLIENT_HEADERS);
2310 if (JB_ERR_OK != err)
2312 assert(err == JB_ERR_PARSE);
2313 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2315 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2317 /* Check request line for rewrites. */
2318 if ((NULL == csp->headers->first->str)
2319 || (strcmp(http->cmd, csp->headers->first->str) &&
2320 (JB_ERR_OK != change_request_destination(csp))))
2323 * A header filter broke the request line - bail out.
2325 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2326 /* XXX: Use correct size */
2327 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2328 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2330 free_http_request(http);
2334 /* decide how to route the HTTP request */
2335 fwd = forward_url(csp, http);
2338 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2339 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2343 * build the http request to send to the server
2344 * we have to do one of the following:
2346 * create = use the original HTTP request to create a new
2347 * HTTP request that has either the path component
2348 * without the http://domainspec (w/path) or the
2349 * full orininal URL (w/url)
2350 * Note that the path and/or the HTTP version may
2351 * have been altered by now.
2353 * connect = Open a socket to the host:port of the server
2354 * and short-circuit server and client socket.
2356 * pass = Pass the request unchanged if forwarding a CONNECT
2357 * request to a parent proxy. Note that we'll be sending
2358 * the CFAIL message ourselves if connecting to the parent
2359 * fails, but we won't send a CSUCCEED message if it works,
2360 * since that would result in a double message (ours and the
2361 * parent's). After sending the request to the parent, we simply
2364 * here's the matrix:
2367 * +--------+--------+
2369 * 0 | create | connect|
2371 * Forwarding +--------+--------+
2373 * 1 | create | pass |
2375 * +--------+--------+
2379 if (http->ssl && connect_port_is_forbidden(csp))
2381 const char *acceptable_connect_ports =
2382 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2383 assert(NULL != acceptable_connect_ports);
2384 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2385 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2386 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2387 csp->action->flags |= ACTION_BLOCK;
2393 freez(csp->headers->first->str);
2394 build_request_line(csp, fwd, &csp->headers->first->str);
2397 hdr = list_to_text(csp->headers);
2400 /* FIXME Should handle error properly */
2401 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2405 * We have a request. Check if one of the crunchers wants it.
2407 if (crunch_response_triggered(csp, crunchers_all))
2410 * Yes. The client got the crunch response
2411 * and we are done here after cleaning up.
2414 list_remove_all(csp->headers);
2420 * The headers can't be removed earlier because
2421 * they were still needed for the referrer check
2422 * in case of CGI crunches.
2424 * XXX: Would it be worth to move the referrer check
2425 * into client_referrer() and set a flag if it's trusted?
2427 list_remove_all(csp->headers);
2429 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2431 if (fwd->forward_host)
2433 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2434 fwd->forward_host, fwd->forward_port, http->hostport);
2438 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2441 /* here we connect to the server, gateway, or the forwarder */
2443 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2444 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2446 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2447 forwarded_connect_retries, http->hostport);
2450 if (csp->sfd == JB_INVALID_SOCKET)
2452 if (fwd->type != SOCKS_NONE)
2455 rsp = error_response(csp, "forwarding-failed", errno);
2457 else if (errno == EINVAL)
2459 rsp = error_response(csp, "no-such-domain", errno);
2463 rsp = error_response(csp, "connect-failed", errno);
2464 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2468 /* Write the answer to the client */
2471 send_crunch_response(csp, rsp);
2478 if (fwd->forward_host || (http->ssl == 0))
2481 * Write the client's (modified) header to the server
2482 * (along with anything else that may be in the buffer)
2484 if (write_socket(csp->sfd, hdr, strlen(hdr))
2485 || (flush_socket(csp->sfd, csp->iob) < 0))
2487 log_error(LOG_LEVEL_CONNECT,
2488 "write header to: %s failed: %E", http->hostport);
2490 rsp = error_response(csp, "connect-failed", errno);
2493 send_crunch_response(csp, rsp);
2503 * We're running an SSL tunnel and we're not forwarding,
2504 * so just send the "connect succeeded" message to the
2505 * client, flush the rest, and get out of the way.
2507 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2515 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2517 /* we're finished with the client's header */
2520 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2522 /* pass data between the client and server
2523 * until one or the other shuts down the connection.
2532 * FD_ZERO here seems to point to an errant macro which crashes.
2533 * So do this by hand for now...
2535 memset(&rfds,0x00,sizeof(fd_set));
2539 FD_SET(csp->cfd, &rfds);
2540 FD_SET(csp->sfd, &rfds);
2542 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2543 if (server_body && server_response_is_complete(csp, byte_count))
2545 log_error(LOG_LEVEL_CONNECT,
2546 "Done reading from server. Expected content length: %d. "
2547 "Actual content length: %d. Most recently received: %d.",
2548 csp->expected_content_length, byte_count, len);
2551 * XXX: should not jump around,
2552 * chat() is complicated enough already.
2556 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2558 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2562 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2567 * This is the body of the browser's request,
2568 * just read and write it.
2570 if (FD_ISSET(csp->cfd, &rfds))
2572 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2576 break; /* "game over, man" */
2579 if (write_socket(csp->sfd, buf, (size_t)len))
2581 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2588 * The server wants to talk. It could be the header or the body.
2589 * If `hdr' is null, then it's the header otherwise it's the body.
2590 * FIXME: Does `hdr' really mean `host'? No.
2592 if (FD_ISSET(csp->sfd, &rfds))
2595 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2599 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2601 if (http->ssl && (fwd->forward_host == NULL))
2604 * Just hang up. We already confirmed the client's CONNECT
2605 * request with status code 200 and unencrypted content is
2606 * no longer welcome.
2608 log_error(LOG_LEVEL_ERROR,
2609 "CONNECT already confirmed. Unable to tell the client about the problem.");
2612 else if (byte_count)
2615 * Just hang up. We already transmitted the original headers
2616 * and parts of the original content and therefore missed the
2617 * chance to send an error message (without risking data corruption).
2619 * XXX: we could retry with a fancy range request here.
2621 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2622 "Unable to tell the client about the problem.");
2626 rsp = error_response(csp, "connect-failed", errno);
2629 send_crunch_response(csp, rsp);
2635 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2636 if (csp->flags & CSP_FLAG_CHUNKED)
2638 if ((len > 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
2640 /* XXX: this is a temporary hack */
2641 log_error(LOG_LEVEL_CONNECT,
2642 "Looks like we reached the end of the last chunk. "
2643 "We better stop reading.");
2644 csp->expected_content_length = byte_count + (size_t)len;
2645 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2649 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2652 * Add a trailing zero to let be able to use string operations.
2653 * XXX: do we still need this with filter_popups gone?
2658 * Normally, this would indicate that we've read
2659 * as much as the server has sent us and we can
2660 * close the client connection. However, Microsoft
2661 * in its wisdom has released IIS/5 with a bug that
2662 * prevents it from sending the trailing \r\n in
2663 * a 302 redirect header (and possibly other headers).
2664 * To work around this if we've haven't parsed
2665 * a full header we'll append a trailing \r\n
2666 * and see if this now generates a valid one.
2668 * This hack shouldn't have any impacts. If we've
2669 * already transmitted the header or if this is a
2670 * SSL connection, then we won't bother with this
2671 * hack. So we only work on partially received
2672 * headers. If we append a \r\n and this still
2673 * doesn't generate a valid header, then we won't
2674 * transmit anything to the client.
2679 if (server_body || http->ssl)
2682 * If we have been buffering up the document,
2683 * now is the time to apply content modification
2684 * and send the result to the client.
2688 p = execute_content_filter(csp, content_filter);
2690 * If the content filter fails, use the original
2691 * buffer and length.
2692 * (see p != NULL ? p : csp->iob->cur below)
2696 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2699 if (JB_ERR_OK != update_server_headers(csp))
2701 log_error(LOG_LEVEL_FATAL,
2702 "Failed to update server headers. after filtering.");
2705 hdr = list_to_text(csp->headers);
2708 /* FIXME Should handle error properly */
2709 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2712 if (write_socket(csp->cfd, hdr, strlen(hdr))
2713 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2715 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2725 break; /* "game over, man" */
2729 * This is NOT the body, so
2730 * Let's pretend the server just sent us a blank line.
2732 snprintf(buf, sizeof(buf), "\r\n");
2733 len = (int)strlen(buf);
2736 * Now, let the normal header parsing algorithm below do its
2737 * job. If it fails, we'll exit instead of continuing.
2744 * If this is an SSL connection or we're in the body
2745 * of the server document, just write it to the client,
2746 * unless we need to buffer the body for later content-filtering
2748 if (server_body || http->ssl)
2753 * If there is no memory left for buffering the content, or the buffer limit
2754 * has been reached, switch to non-filtering mode, i.e. make & write the
2755 * header, flush the iob and buf, and get out of the way.
2757 if (add_to_iob(csp, buf, len))
2762 log_error(LOG_LEVEL_INFO,
2763 "Flushing header and buffers. Stepping back from filtering.");
2765 hdr = list_to_text(csp->headers);
2769 * Memory is too tight to even generate the header.
2770 * Send our static "Out-of-memory" page.
2772 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2773 rsp = cgi_error_memory();
2774 send_crunch_response(csp, rsp);
2777 hdrlen = strlen(hdr);
2779 if (write_socket(csp->cfd, hdr, hdrlen)
2780 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2781 || (write_socket(csp->cfd, buf, (size_t)len)))
2783 log_error(LOG_LEVEL_CONNECT,
2784 "Flush header and buffers to client failed: %E");
2790 * Reset the byte_count to the amount of bytes
2791 * we just flushed. len will be added a few lines below,
2792 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2794 byte_count = (size_t)flushed;
2796 content_filter = NULL;
2802 if (write_socket(csp->cfd, buf, (size_t)len))
2804 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2808 byte_count += (size_t)len;
2813 const char *header_start;
2815 * We're still looking for the end of the server's header.
2816 * Buffer up the data we just read. If that fails, there's
2817 * little we can do but send our static out-of-memory page.
2819 if (add_to_iob(csp, buf, len))
2821 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2822 rsp = cgi_error_memory();
2823 send_crunch_response(csp, rsp);
2827 header_start = csp->iob->cur;
2829 /* Convert iob into something sed() can digest */
2830 if (JB_ERR_PARSE == get_server_headers(csp))
2835 * Well, we tried our MS IIS/5 hack and it didn't work.
2836 * The header is incomplete and there isn't anything
2837 * we can do about it.
2839 log_error(LOG_LEVEL_INFO,
2840 "MS IIS5 hack didn't produce valid headers.");
2846 * Since we have to wait for more from the server before
2847 * we can parse the headers we just continue here.
2853 /* Did we actually get anything? */
2854 if (NULL == csp->headers->first)
2856 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2857 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2858 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2859 free_http_request(http);
2863 assert(csp->headers->first->str);
2865 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2866 strncmpic(csp->headers->first->str, "ICY", 3))
2869 * It doesn't look like a HTTP (or Shoutcast) response:
2870 * tell the client and log the problem.
2872 if (strlen(csp->headers->first->str) > 30)
2874 csp->headers->first->str[30] = '\0';
2876 log_error(LOG_LEVEL_ERROR,
2877 "Invalid server or forwarder response. Starts with: %s",
2878 csp->headers->first->str);
2879 log_error(LOG_LEVEL_CLF,
2880 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2881 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2882 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2883 free_http_request(http);
2888 * We have now received the entire server header,
2889 * filter it and send the result to the client
2891 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2893 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2895 hdr = list_to_text(csp->headers);
2898 /* FIXME Should handle error properly */
2899 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2902 if (crunch_response_triggered(csp, crunchers_light))
2905 * One of the tags created by a server-header
2906 * tagger triggered a crunch. We already
2907 * delivered the crunch response to the client
2908 * and are done here after cleaning up.
2913 /* Buffer and pcrs filter this if appropriate. */
2915 if (!http->ssl) /* We talk plaintext */
2917 content_filter = get_filter_function(csp);
2920 * Only write if we're not buffering for content modification
2922 if (!content_filter)
2925 * Write the server's (modified) header to
2926 * the client (along with anything else that
2927 * may be in the buffer)
2930 if (write_socket(csp->cfd, hdr, strlen(hdr))
2931 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2933 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2936 * The write failed, so don't bother mentioning it
2937 * to the client... it probably can't hear us anyway.
2943 byte_count += (size_t)len;
2948 * XXX: the header lenght should probably
2949 * be calculated by get_server_headers().
2951 int header_length = csp->iob->cur - header_start;
2952 assert(csp->iob->cur > header_start);
2953 byte_count += (size_t)len - header_length;
2956 /* we're finished with the server's header */
2962 * If this was a MS IIS/5 hack then it means the server
2963 * has already closed the connection. Nothing more to read.
2968 log_error(LOG_LEVEL_INFO,
2969 "Closed server connection detected with MS IIS5 hack enabled.");
2976 * If we reach this point, the server socket is tainted
2977 * (most likely because we didn't read everything the
2978 * server sent us) and reusing it would lead to garbage.
2980 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2982 log_error(LOG_LEVEL_CONNECT, "Unsetting keep-alive flag.");
2983 csp->flags &= ~CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE;
2988 if (csp->content_length == 0)
2991 * If Privoxy didn't recalculate the Content-Lenght,
2992 * byte_count is still correct.
2994 csp->content_length = byte_count;
2997 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2998 csp->ip_addr_str, http->ocmd, csp->content_length);
3002 /*********************************************************************
3006 * Description : This is little more than chat. We only "serve" to
3007 * to close any socket that chat may have opened.
3010 * 1 : csp = Current client state (buffers, headers, etc...)
3014 *********************************************************************/
3016 void serve(struct client_state *csp)
3017 #else /* ifndef AMIGA */
3018 static void serve(struct client_state *csp)
3019 #endif /* def AMIGA */
3022 close_socket(csp->cfd);
3024 if (csp->sfd != JB_INVALID_SOCKET)
3026 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3027 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
3029 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
3033 forget_connection(csp->sfd);
3034 close_socket(csp->sfd);
3037 close_socket(csp->sfd);
3038 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3041 csp->flags &= ~CSP_FLAG_ACTIVE;
3047 /*********************************************************************
3049 * Function : server_thread
3051 * Description : We only exist to call `serve' in a threaded environment.
3054 * 1 : data = Current client state (buffers, headers, etc...)
3056 * Returns : Always 0.
3058 *********************************************************************/
3059 static int32 server_thread(void *data)
3061 serve((struct client_state *) data);
3068 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3069 /*********************************************************************
3073 * Description : Print usage info & exit.
3075 * Parameters : Pointer to argv[0] for identifying ourselves
3079 *********************************************************************/
3080 static void usage(const char *myname)
3082 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3086 #endif /* defined(unix) */
3089 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3090 #endif /* defined(unix) */
3091 "[--version] [configfile]\n"
3092 "Aborting\n", myname);
3097 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3100 #ifdef MUTEX_LOCKS_AVAILABLE
3101 /*********************************************************************
3103 * Function : privoxy_mutex_lock
3105 * Description : Locks a mutex.
3108 * 1 : mutex = The mutex to lock.
3110 * Returns : Void. May exit in case of errors.
3112 *********************************************************************/
3113 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3115 #ifdef FEATURE_PTHREAD
3116 int err = pthread_mutex_lock(mutex);
3119 if (mutex != &log_mutex)
3121 log_error(LOG_LEVEL_FATAL,
3122 "Mutex locking failed: %s.\n", strerror(err));
3127 EnterCriticalSection(mutex);
3128 #endif /* def FEATURE_PTHREAD */
3132 /*********************************************************************
3134 * Function : privoxy_mutex_unlock
3136 * Description : Unlocks a mutex.
3139 * 1 : mutex = The mutex to unlock.
3141 * Returns : Void. May exit in case of errors.
3143 *********************************************************************/
3144 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3146 #ifdef FEATURE_PTHREAD
3147 int err = pthread_mutex_unlock(mutex);
3150 if (mutex != &log_mutex)
3152 log_error(LOG_LEVEL_FATAL,
3153 "Mutex unlocking failed: %s.\n", strerror(err));
3158 LeaveCriticalSection(mutex);
3159 #endif /* def FEATURE_PTHREAD */
3163 /*********************************************************************
3165 * Function : privoxy_mutex_init
3167 * Description : Prepares a mutex.
3170 * 1 : mutex = The mutex to initialize.
3172 * Returns : Void. May exit in case of errors.
3174 *********************************************************************/
3175 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3177 #ifdef FEATURE_PTHREAD
3178 int err = pthread_mutex_init(mutex, 0);
3181 printf("Fatal error. Mutex initialization failed: %s.\n",
3186 InitializeCriticalSection(mutex);
3187 #endif /* def FEATURE_PTHREAD */
3189 #endif /* def MUTEX_LOCKS_AVAILABLE */
3191 /*********************************************************************
3193 * Function : initialize_mutexes
3195 * Description : Prepares mutexes if mutex support is available.
3199 * Returns : Void, exits in case of errors.
3201 *********************************************************************/
3202 static void initialize_mutexes(void)
3204 #ifdef MUTEX_LOCKS_AVAILABLE
3206 * Prepare global mutex semaphores
3208 privoxy_mutex_init(&log_mutex);
3209 privoxy_mutex_init(&log_init_mutex);
3210 privoxy_mutex_init(&connection_reuse_mutex);
3213 * XXX: The assumptions below are a bit naive
3214 * and can cause locks that aren't necessary.
3216 * For example older FreeBSD versions (< 6.x?)
3217 * have no gethostbyname_r, but gethostbyname is
3220 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3221 privoxy_mutex_init(&resolver_mutex);
3222 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3224 * XXX: should we use a single mutex for
3225 * localtime() and gmtime() as well?
3227 #ifndef HAVE_GMTIME_R
3228 privoxy_mutex_init(&gmtime_mutex);
3229 #endif /* ndef HAVE_GMTIME_R */
3231 #ifndef HAVE_LOCALTIME_R
3232 privoxy_mutex_init(&localtime_mutex);
3233 #endif /* ndef HAVE_GMTIME_R */
3236 privoxy_mutex_init(&rand_mutex);
3237 #endif /* ndef HAVE_RANDOM */
3238 #endif /* def MUTEX_LOCKS_AVAILABLE */
3242 /*********************************************************************
3246 * Description : Load the config file and start the listen loop.
3247 * This function is a lot more *sane* with the `load_config'
3248 * and `listen_loop' functions; although it stills does
3249 * a *little* too much for my taste.
3252 * 1 : argc = Number of parameters (including $0).
3253 * 2 : argv = Array of (char *)'s to the parameters.
3255 * Returns : 1 if : can't open config file, unrecognized directive,
3256 * stats requested in multi-thread mode, can't open the
3257 * log file, can't open the jar file, listen port is invalid,
3258 * any load fails, and can't bind port.
3260 * Else main never returns, the process must be signaled
3261 * to terminate execution. Or, on Windows, use the
3262 * "File", "Exit" menu option.
3264 *********************************************************************/
3266 int real_main(int argc, const char *argv[])
3268 int main(int argc, const char *argv[])
3272 unsigned int random_seed;
3274 struct passwd *pw = NULL;
3275 struct group *grp = NULL;
3278 char *pre_chroot_nslookup_to_load_resolver = NULL;
3285 #if !defined(_WIN32)
3293 * Parse the command line arguments
3295 * XXX: simply printing usage information in case of
3296 * invalid arguments isn't particularly user friendly.
3298 while (++argc_pos < argc)
3301 /* Check to see if the service must be installed or uninstalled */
3302 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3304 const char *pName = argv[argc_pos] + 9;
3307 exit( (install_service(pName)) ? 0 : 1 );
3309 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3311 const char *pName = argv[argc_pos] + 11;
3314 exit((uninstall_service(pName)) ? 0 : 1);
3316 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3318 bRunAsService = TRUE;
3319 w32_set_service_cwd();
3320 atexit(w32_service_exit_notify);
3323 #endif /* defined(_WIN32) */
3326 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3328 if (strcmp(argv[argc_pos], "--help") == 0)
3333 else if(strcmp(argv[argc_pos], "--version") == 0)
3335 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3341 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3346 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3348 if (++argc_pos == argc) usage(argv[0]);
3349 pidfile = strdup(argv[argc_pos]);
3352 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3354 if (++argc_pos == argc) usage(argv[argc_pos]);
3356 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3359 if (NULL == (grp = getgrnam(p)))
3361 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3365 if (NULL == (pw = getpwnam(argv[argc_pos])))
3367 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3370 if (p != NULL) *--p = '\0';
3373 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3375 if (++argc_pos == argc) usage(argv[0]);
3376 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3379 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3383 #endif /* defined(unix) */
3385 else if (argc_pos + 1 != argc)
3388 * This is neither the last command line
3389 * option, nor was it recognized before,
3390 * therefore it must be invalid.
3396 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3398 configfile = argv[argc_pos];
3401 } /* -END- while (more arguments) */
3404 if ( *configfile != '/' )
3406 char cwd[BUFFER_SIZE];
3408 size_t abs_file_size;
3410 /* make config-filename absolute here */
3411 if (NULL == getcwd(cwd, sizeof(cwd)))
3413 perror("failed to get current working directory");
3418 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3419 basedir = strdup(cwd);
3421 if (NULL == basedir ||
3422 NULL == (abs_file = malloc(abs_file_size)))
3424 perror("malloc failed");
3427 strlcpy(abs_file, basedir, abs_file_size);
3428 strlcat(abs_file, "/", abs_file_size );
3429 strlcat(abs_file, configfile, abs_file_size);
3430 configfile = abs_file;
3432 #endif /* defined unix */
3436 clients->next = NULL;
3438 /* XXX: factor out initialising after the next stable release. */
3441 #elif defined(_WIN32)
3445 /* Prepare mutexes if supported and necessary. */
3446 initialize_mutexes();
3448 /* Enable logging until further notice. */
3449 init_log_module(Argv[0]);
3451 random_seed = (unsigned int)time(NULL);
3453 srandom(random_seed);
3456 #endif /* ifdef HAVE_RANDOM */
3459 * Unix signal handling
3461 * Catch the abort, interrupt and terminate signals for a graceful exit
3462 * Catch the hangup signal so the errlog can be reopened.
3463 * Ignore the broken pipe signals (FIXME: Why?)
3465 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3468 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3469 const int ignored_signals[] = { SIGPIPE, 0 };
3471 for (idx = 0; catched_signals[idx] != 0; idx++)
3473 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3474 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3476 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3477 #endif /* ifdef sun */
3479 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3483 for (idx = 0; ignored_signals[idx] != 0; idx++)
3485 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3487 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3492 #else /* ifdef _WIN32 */
3493 # ifdef _WIN_CONSOLE
3495 * We *are* in a windows console app.
3496 * Print a verbose messages about FAQ's and such
3498 printf("%s", win32_blurb);
3499 # endif /* def _WIN_CONSOLE */
3500 #endif /* def _WIN32 */
3503 /* Initialize the CGI subsystem */
3504 cgi_init_error_messages();
3507 * If runnig on unix and without the --nodaemon
3508 * option, become a daemon. I.e. fork, detach
3509 * from tty and get process group leadership
3522 if ( pid < 0 ) /* error */
3527 else if ( pid != 0 ) /* parent */
3532 * must check for errors
3533 * child died due to missing files aso
3536 wpid = waitpid( pid, &status, WNOHANG );
3545 /* Should be more portable, but not as well tested */
3550 #else /* ndef __FreeBSD__ */
3552 #endif /* ndef __FreeBSD__ */
3553 fd = open("/dev/tty", O_RDONLY);
3556 /* no error check here */
3557 ioctl( fd, TIOCNOTTY,0 );
3562 * stderr (fd 2) will be closed later on, when the
3563 * log file has been parsed.
3570 } /* -END- if (!no_daemon) */
3573 * As soon as we have written the PID file, we can switch
3574 * to the user and group ID indicated by the --user option
3580 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3582 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3588 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3590 /* Read the time zone file from /etc before doing chroot. */
3592 if (NULL != pre_chroot_nslookup_to_load_resolver
3593 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3595 /* Initialize resolver library. */
3596 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3598 if (chroot(pw->pw_dir) < 0)
3600 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3604 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3607 if (setuid(pw->pw_uid))
3609 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3613 char putenv_dummy[64];
3615 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3616 if (putenv(putenv_dummy) != 0)
3618 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3621 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3622 if (putenv(putenv_dummy) != 0)
3624 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3630 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3633 #endif /* defined unix */
3636 /* This will be FALSE unless the command line specified --service
3640 /* Yup, so now we must attempt to establish a connection
3641 * with the service dispatcher. This will only work if this
3642 * process was launched by the service control manager to
3643 * actually run as a service. If this isn't the case, i've
3644 * known it take around 30 seconds or so for the call to return.
3647 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3648 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3650 /* Service has run, and at this point is now being stopped, so just return */
3655 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3657 /* An error occurred. Usually it's because --service was wrongly specified
3658 * and we were unable to connect to the Service Control Dispatcher because
3659 * it wasn't expecting us and is therefore not listening.
3661 * For now, just continue below to call the listen_loop function.
3664 #endif /* def _WIN32 */
3674 /*********************************************************************
3676 * Function : bind_port_helper
3678 * Description : Bind the listen port. Handles logging, and aborts
3682 * 1 : config = Privoxy configuration. Specifies port
3685 * Returns : Port that was opened.
3687 *********************************************************************/
3688 static jb_socket bind_port_helper(struct configuration_spec * config)
3693 if (config->haddr == NULL)
3695 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3700 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3701 config->hport, config->haddr);
3704 result = bind_port(config->haddr, config->hport, &bfd);
3711 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3712 "There may be another Privoxy or some other "
3713 "proxy running on port %d",
3714 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3715 config->hport, config->hport);
3718 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3719 "The hostname is not resolvable",
3720 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3723 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3724 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3727 /* shouldn't get here */
3728 return JB_INVALID_SOCKET;
3731 config->need_bind = 0;
3738 /* Without this simple workaround we get this compiler warning from _beginthread
3739 * warning C4028: formal parameter 1 different from declaration
3741 void w32_service_listen_loop(void *p)
3745 #endif /* def _WIN32 */
3748 /*********************************************************************
3750 * Function : listen_loop
3752 * Description : bind the listen port and enter a "FOREVER" listening loop.
3758 *********************************************************************/
3759 static void listen_loop(void)
3761 struct client_state *csp = NULL;
3763 struct configuration_spec * config;
3765 config = load_config();
3767 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3769 * XXX: Should be relocated once it no
3770 * longer needs to emit log messages.
3772 initialize_reusable_connections();
3773 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3775 bfd = bind_port_helper(config);
3777 #ifdef FEATURE_GRACEFUL_TERMINATION
3778 while (!g_terminate)
3783 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3784 while (waitpid(-1, NULL, WNOHANG) > 0)
3786 /* zombie children */
3788 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3791 * Free data that was used by died threads
3797 * Re-open the errlog after HUP signal
3799 if (received_hup_signal)
3801 init_error_log(Argv[0], config->logfile);
3802 received_hup_signal = 0;
3806 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3808 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3812 csp->flags |= CSP_FLAG_ACTIVE;
3813 csp->sfd = JB_INVALID_SOCKET;
3815 csp->config = config = load_config();
3817 if ( config->need_bind )
3820 * Since we were listening to the "old port", we will not see
3821 * a "listen" param change until the next IJB request. So, at
3822 * least 1 more request must be made for us to find the new
3823 * setting. I am simply closing the old socket and binding the
3826 * Which-ever is correct, we will serve 1 more page via the
3827 * old settings. This should probably be a "show-proxy-args"
3828 * request. This should not be a so common of an operation
3829 * that this will hurt people's feelings.
3834 bfd = bind_port_helper(config);
3837 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3839 if (!accept_connection(csp, bfd))
3841 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3854 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3857 #ifdef FEATURE_TOGGLE
3858 if (global_toggle_state)
3859 #endif /* def FEATURE_TOGGLE */
3861 csp->flags |= CSP_FLAG_TOGGLED_ON;
3864 if (run_loader(csp))
3866 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3867 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3871 if (block_acl(NULL,csp))
3873 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3874 close_socket(csp->cfd);
3878 #endif /* def FEATURE_ACL */
3880 /* add it to the list of clients */
3881 csp->next = clients->next;
3882 clients->next = csp;
3884 if (config->multi_threaded)
3888 /* this is a switch () statment in the C preprocessor - ugh */
3889 #undef SELECTED_ONE_OPTION
3891 /* Use Pthreads in preference to native code */
3892 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3893 #define SELECTED_ONE_OPTION
3895 pthread_t the_thread;
3896 pthread_attr_t attrs;
3898 pthread_attr_init(&attrs);
3899 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3900 errno = pthread_create(&the_thread, &attrs,
3901 (void * (*)(void *))serve, csp);
3902 child_id = errno ? -1 : 0;
3903 pthread_attr_destroy(&attrs);
3907 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3908 #define SELECTED_ONE_OPTION
3909 child_id = _beginthread(
3910 (void (*)(void *))serve,
3915 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3916 #define SELECTED_ONE_OPTION
3917 child_id = _beginthread(
3918 (void(* _Optlink)(void*))serve,
3924 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3925 #define SELECTED_ONE_OPTION
3927 thread_id tid = spawn_thread
3928 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3930 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3932 child_id = (int) tid;
3941 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3942 #define SELECTED_ONE_OPTION
3943 csp->cfd = ReleaseSocket(csp->cfd, -1);
3946 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3947 NP_Output, Output(),
3948 NP_CloseOutput, FALSE,
3949 NP_Name, (ULONG)"privoxy child",
3953 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3954 NP_Output, Output(),
3955 NP_CloseOutput, FALSE,
3956 NP_Name, (ULONG)"privoxy child",
3957 NP_StackSize, 200*1024,
3963 ((struct Task *)child_id)->tc_UserData = csp;
3964 Signal((struct Task *)child_id, SIGF_SINGLE);
3969 #if !defined(SELECTED_ONE_OPTION)
3972 /* This block is only needed when using fork().
3973 * When using threads, the server thread was
3974 * created and run by the call to _beginthread().
3976 if (child_id == 0) /* child */
3979 #ifdef FEATURE_TOGGLE
3980 int inherited_toggle_state = global_toggle_state;
3981 #endif /* def FEATURE_TOGGLE */
3986 * If we've been toggled or we've blocked the request, tell Mom
3989 #ifdef FEATURE_TOGGLE
3990 if (inherited_toggle_state != global_toggle_state)
3992 rc |= RC_FLAG_TOGGLED;
3994 #endif /* def FEATURE_TOGGLE */
3996 #ifdef FEATURE_STATISTICS
3997 if (csp->flags & CSP_FLAG_REJECTED)
3999 rc |= RC_FLAG_BLOCKED;
4001 #endif /* ndef FEATURE_STATISTICS */
4005 else if (child_id > 0) /* parent */
4007 /* in a fork()'d environment, the parent's
4008 * copy of the client socket and the CSP
4012 #if !defined(_WIN32) && !defined(__CYGWIN__)
4014 wait( &child_status );
4017 * Evaluate child's return code: If the child has
4018 * - been toggled, toggle ourselves
4019 * - blocked its request, bump up the stats counter
4022 #ifdef FEATURE_TOGGLE
4023 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
4025 global_toggle_state = !global_toggle_state;
4027 #endif /* def FEATURE_TOGGLE */
4029 #ifdef FEATURE_STATISTICS
4031 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4035 #endif /* def FEATURE_STATISTICS */
4037 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4038 close_socket(csp->cfd);
4039 csp->flags &= ~CSP_FLAG_ACTIVE;
4043 #undef SELECTED_ONE_OPTION
4044 /* end of cpp switch () */
4046 if (child_id < 0) /* failed */
4048 char buf[BUFFER_SIZE];
4050 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
4052 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
4054 write_socket(csp->cfd, buf, strlen(buf));
4055 close_socket(csp->cfd);
4056 csp->flags &= ~CSP_FLAG_ACTIVE;
4067 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4069 /* Clean up. Aim: free all memory (no leaks) */
4070 #ifdef FEATURE_GRACEFUL_TERMINATION
4072 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4074 unload_current_config_file();
4075 unload_current_actions_file();
4076 unload_current_re_filterfile();
4077 #ifdef FEATURE_TRUST
4078 unload_current_trust_file();
4081 if (config->multi_threaded)
4088 } while ((clients->next != NULL) && (--i > 0));
4092 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4103 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4104 /* Cleanup - remove taskbar icon etc. */
4109 #endif /* FEATURE_GRACEFUL_TERMINATION */