1 const char jcc_rcs[] = "$Id: jcc.c,v 1.191 2008/10/11 18:00:14 fabiankeil Exp $";
2 /*********************************************************************
4 * File : $Source: /cvsroot/ijbswa/current/jcc.c,v $
6 * Purpose : Main file. Contains main() method, main loop, and
7 * the main connection-handling function.
9 * Copyright : Written by and Copyright (C) 2001-2008 the SourceForge
10 * Privoxy team. http://www.privoxy.org/
12 * Based on the Internet Junkbuster originally written
13 * by and Copyright (C) 1997 Anonymous Coders and
14 * Junkbusters Corporation. http://www.junkbusters.com
16 * This program is free software; you can redistribute it
17 * and/or modify it under the terms of the GNU General
18 * Public License as published by the Free Software
19 * Foundation; either version 2 of the License, or (at
20 * your option) any later version.
22 * This program is distributed in the hope that it will
23 * be useful, but WITHOUT ANY WARRANTY; without even the
24 * implied warranty of MERCHANTABILITY or FITNESS FOR A
25 * PARTICULAR PURPOSE. See the GNU General Public
26 * License for more details.
28 * The GNU General Public License should be included with
29 * this file. If not, you can view it at
30 * http://www.gnu.org/copyleft/gpl.html
31 * or write to the Free Software Foundation, Inc., 59
32 * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
36 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
37 * Reformat some comments in chat().
39 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
40 * In case of chunk-encoded content, stop reading if
41 * the buffer looks like it ends with the last chunk.
43 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
44 * Let server_response_is_complete() deal properly with
45 * content that is neither buffered nor read all at once.
47 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
48 * Flush work-in-progress changes to keep outgoing connections
49 * alive where possible. Incomplete and mostly #ifdef'd out.
51 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
52 * Add mutex lock support for _WIN32.
54 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
55 * Prepare for critical sections on Windows by adding a
56 * layer of indirection before the pthread mutex functions.
58 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
59 * Remove FEATURE_COOKIE_JAR.
61 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
62 * - Silence LLVM/Clang complaint.
63 * - Make received_hup_signal static.
64 * - Hide definitions for basedir, pidfile and received_hup_signal
65 * from __EMX__ as they only seem to be used in case of #ifdef unix.
67 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
68 * Accept Shoutcast responses again. Problem reported
69 * and fix suggested by Stefan in #2062860.
71 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
72 * Fix possible NULL-pointer dereference reported
73 * by din_a4 in #2003937. Pointy hat to me.
75 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
76 * Streamline sed()'s prototype and declare
77 * the header parse and add structures static.
79 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
80 * - Mark csp as immutable for send_crunch_response().
81 * - Fix comment spelling.
83 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
84 * Factor update_server_headers() out of sed(), ditch the
85 * first_run hack and make server_patterns_light static.
87 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
88 * Don't provide get_header() with the whole client state
89 * structure when it only needs access to csp->iob.
91 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
92 * Make the "read the rest of the headers" loop a bit more readable.
94 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
95 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
96 * - Remove useless comment.
98 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
99 * Fix comment grammar.
101 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
102 * Remove the pointless buffer in client_protocol_is_unsupported().
104 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
105 * Least-effort fix for bug #1821930 (reported by Lee):
106 * If the response doesn't look like HTTP,
107 * tell the client and log the problem.
109 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
110 * Don't pass the whole csp structure to flush_socket()
111 * when it only needs a file descriptor and a buffer.
113 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
114 * Remove kill-popups action.
116 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
117 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
119 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
120 * Remove the treat-forbidden-connects-like-blocks action. We now
121 * use the "blocked" page for forbidden CONNECT requests by default.
123 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
124 * Also use shiny new connect_port_is_forbidden() in jcc.c.
126 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
127 * Rename url_actions() to get_url_actions() and let it
128 * use the standard parameter ordering.
130 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
131 * Let forward_url() use the standard parameter ordering
132 * and mark its second parameter immutable.
134 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
135 * Remove the "Listening ... for local connections only" log message.
136 * Whether or not remote connections are able to reach Privoxy is up
137 * to the operating system.
139 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
140 * Prevent the log messages for CONNECT requests to unacceptable
141 * ports from printing the limit-connect argument as [null] if
142 * limit-connect hasn't been explicitly enabled.
144 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
145 * Make sure all console-mode apps get a usage() instance
147 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
148 * Reword NO_SERVER_DATA_RESPONSE to make it harder
149 * to misunderstand what the message is all about.
151 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
152 * Unbreak trustfile which previously didn't work without
153 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
155 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
156 * Plug memory leak. Spotted by Valgrind, triggered by
157 * Privoxy-Regression-Test feeding proxyfuzz.py.
159 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
160 * In the HTTP snipplets, refer to the client as client.
162 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
163 * Emit a log message when activating the MS IIS5 hack.
165 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
166 * Log the "weak randomization factor" warning only
167 * once for mingw32 and provide some more details.
169 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
170 * Initialize log module after initializing mutexes, future
171 * deadlocks in that code should now work cross-platform.
173 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
174 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
175 * Reported by Bert van Leeuwen in BR#1818808.
177 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
178 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
180 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
181 * When in daemon mode, close stderr after the configuration file has been
182 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
184 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
185 * - Fix a crash when parsing invalid requests whose first header
186 * is rejected by get_header(). Regression (re?)introduced
187 * in r1.143 by yours truly.
188 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
189 * client_cookie_adder() to make sure send-vanilla-wafer can be
190 * controlled through tags (and thus regression-tested).
192 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
193 * - Move get_filter_function() from jcc.c to filters.c
194 * so the filter functions can be static.
195 * - Don't bother filtering body-less responses.
197 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
198 * Execute content filters through execute_content_filter().
200 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
201 * Initialize req to NULL to make sure it's defined if the
202 * first read_socket() call fails. Reported by icmp30.
204 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
205 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
206 * extensive comments moved to user manual.
208 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
209 * Don't crash if a broken header filter wiped out the request line.
211 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
212 * Fix byte_count calculation in case of flushes
213 * and don't parse the server headers a second time.
215 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
216 * - If there's a connection problem after we already forwarded
217 * parts of the original content, just hang up. Fixes BR#1776724.
218 * - Fix warnings about unused code on mingw32.
219 * - In case of flushes, calculate the byte count
220 * less incorrectly (I think).
222 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
223 * Add some more prototypes for static functions.
225 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
226 * Comment out request_contains_null_bytes() until it's used again.
228 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
229 * #1763173 from Stefan Huehner: s@const static@static const@
230 * and declare some more functions static.
232 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
233 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
234 * and explain why they were rejected in the first place.
235 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
236 * CONNECT requests. The request line was missing.
237 * - Add two more XXX reminders as we don't have enough already.
239 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
240 * As Hal noticed, checking dispatch_cgi() as the last cruncher
241 * looks like a bug if CGI requests are blocked unintentionally,
242 * so don't do it unless the user enabled the new config option
243 * "allow-cgi-request-crunching".
245 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
246 * - Allow to rewrite the request destination behind the client's back.
247 * - Turn the weird-looking unconditional for loop that
248 * reads the client request into a conditional while loop.
249 * Move the stuff that only runs once out of the loop.
250 * - Move parts of chat(), server_content_type() and the
251 * necessary stuff to fix BR#1750917 into get_filter_function().
253 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
254 * Temporary workaround for BR#1730105.
256 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
257 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
258 * deadlocks and crashes on OpenBSD and possibly other OS with neither
259 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
260 * Thanks to Ralf Horstmann for report and solution.
262 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
263 * Add forward-override{} to change the forwarding settings through
264 * action sections. This is mainly interesting to forward different
265 * clients differently (for example based on User-Agent or request
268 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
269 * - Let usage() mention the --chroot parameter.
270 * - Use read_socket() consistently and always leave
271 * the last buffer byte alone, even in cases where
272 * null termination (currently) doesn't matter.
274 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
275 * - Fix config file loading on Unix if no config file is specified.
276 * Since r1.97 Privoxy would always interpret the last argument as
277 * config file, even if it's a valid command line option.
278 * - Abort in case of unrecognized command line options. Closes #1719696.
279 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
280 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
282 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
283 * - Don't rerun crunchers that only depend on the request URL.
284 * - Don't count redirects and CGI requests as "blocked requests".
286 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
287 * Support crunching based on tags created by server-header taggers.
289 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
290 * Make HTTP snippets static (again). Add a Content-Type for those
291 * with content so the browser doesn't guess it based on the URL.
293 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
294 * Move crunching and request line rebuilding out of chat().
296 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
297 * Introduce tags as alternative way to specify which
298 * actions apply to a request. At the moment tags can be
299 * created based on client and server headers.
301 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
302 * Don't CLF-log CONNECT requests twice.
304 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
305 * Log the source address for ACL-related connection drops.
307 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
308 * New config option: enforce-blocks.
310 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
311 * - Move null byte check into separate function.
312 * - Don't confuse the client with error pages
313 * if a CONNECT request was already confirmed.
315 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
316 * Speed up NULL byte escaping and only log the complete
317 * NULL byte requests with header debugging enabled.
319 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
320 * Answer requests that contain NULL bytes with
321 * a custom response instead of waiting for more
322 * data until the client eventually hangs up.
324 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
325 * - Move delivery and logging of crunched responses
326 * from chat() into send_crunch_response().
327 * - Display the reason for generating http_responses.
328 * - Log the content length for LOG_LEVEL_CLF correctly
329 * (still incorrect for some fixed responses).
330 * - Reword an incorrect comment about
331 * treat-forbidden-connects-like-blocks violating
333 * - Add some log messages.
335 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
336 * Move mutex initialization into separate
337 * function and exit in case of errors.
339 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
340 * - Start to reduce chat()'s line count and move
341 * parts of it into separate functions.
342 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
344 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
345 * - Add Proxy-Agent header to HTTP snippets that are
346 * supposed to reach HTTP clients only.
347 * - Made a few CONNECT log messages more descriptive.
348 * - Catch completely empty server responses (as seen
349 * with Tor's fake ".noconnect" top level domain).
350 * - Use shiny new "forwarding-failed" template for socks errors.
352 * Revision 1.118 2007/01/07 07:43:43 joergs
353 * AmigaOS4 support added.
355 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
356 * Added config option accept-intercepted-requests
357 * and disabled it by default.
359 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
360 * Reverted parts of my last commit
361 * to keep error handling working.
363 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
364 * Fixed gcc43 conversion warnings.
366 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
367 * Fix -pedantic ISO C warning about converting
368 * from function pointer to object pointer.
370 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
371 * Silence compiler warning I introduced with my last commit.
373 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
374 * Mutex protect rand() if POSIX threading
375 * is used, warn the user if that's not possible
376 * and stop using it on _WIN32 where it could
379 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
380 * Don't prevent core dumps by catching SIGABRT.
381 * It's rude and makes debugging unreasonable painful.
383 * Revision 1.110 2006/12/13 14:52:53 etresoft
384 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
386 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
387 * Privoxy is now able to run as intercepting
388 * proxy in combination with any packet filter
389 * that does the port redirection. The destination
390 * is extracted from the "Host:" header which
391 * should be available for nearly all requests.
393 * Moved HTTP snipplets into jcc.c.
394 * Added error message for gopher proxy requests.
396 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
397 * Only unlink the pidfile if it's actually used.
399 * Change order of interception checks to make
400 * it possible to block or redirect requests for
403 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
404 * Make pthread mutex locking more generic. Instead of
405 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
406 * and use mutex locking unless there is an _r function
407 * available. Better safe than sorry.
409 * Fixes "./configure --disable-pthread" and should result
410 * in less threading-related problems on pthread-using platforms,
411 * but it still doesn't fix BR#1122404.
413 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
414 * Move pthread.h inclusion from jcc.c to jcc.h.
415 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
417 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
418 * Don't exit after receiving the second SIGHUP on Solaris.
420 * Fixes BR 1052235, but the same problem may exist on other
421 * systems. Once 3.0.6 is out we should use sigset()
422 * where available and see if it breaks anything.
424 * Revision 1.104 2006/09/23 13:26:38 roro
425 * Replace TABs by spaces in source code.
427 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
428 * Fix +redirect{}. Didn't work with -fast-redirects.
430 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
431 * Respond with 400 and a short text message
432 * if the client tries to use Privoxy as FTP proxy.
434 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
435 * Make number of retries in case of forwarded-connect problems
436 * a config file option (forwarded-connect-retries) and use 0 as
439 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
440 * Set random(3) seed.
442 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
443 * Follow the OpenBSD port's lead and protect the resolve
444 * functions on OpenBSD as well.
446 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
447 * --user fix. Only use the user as group if no group is specified.
448 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
450 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
451 * Windows service (re-)integration
455 * --install[:service_name]
456 * --uninstall[:service_name]
459 * They work as follows:
460 * --install will create a service for you and then terminate.
461 * By default the service name will be "privoxy" (without the quotes).
462 * However you can run multiple services if you wish, just by adding
463 * a colon and then a name (no spaces).
465 * --uninstall follows the exact same rules a --install.
467 * --service is used when the program is executed by the service
468 * control manager, and in normal circumstances would never be
469 * used as a command line argument.
471 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
472 * Windows service integration
474 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
475 * Incorporate Fabian Keil's patch work:
476 http://www.fabiankeil.de/sourcecode/privoxy/
478 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
479 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
480 * with what was really the latest development (the v_3_0_branch branch)
482 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
483 * Thanks to Jindrich Makovicka for a race condition fix for the log
484 * file. The race condition remains for non-pthread implementations.
485 * Reference patch #1175720.
487 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
488 * Add the ability to check jpeg images for invalid
489 * lengths of comment blocks. Defensive strategy
490 * against the exploit:
491 * Microsoft Security Bulletin MS04-028
492 * Buffer Overrun in JPEG Processing (GDI+) Could
493 * Allow Code Execution (833987)
494 * Enabled with +inspect-jpegs in actions files.
496 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
497 * - Fixed usage info for non-unix platforms
498 * - Fixed small cmdline parsing bug
500 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
501 * Diagnostics: Now preserve the returncode of pthread_create
502 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
504 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
505 * No longer ignore SIGCHLD. Fixes bug #769381
507 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
508 * Close jarfile on graceful exit, remove stray line
510 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
511 * Cosmetics: Killed a warning, a typo and an allocation left at exit
513 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
514 * No longer rely on non-POSIX.1 extensions of getcwd().
517 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
518 * Replaced setenv() by posix-compliant putenv()
519 * Thanks to Neil McCalden (nmcc AT users.sf.net).
521 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
522 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
524 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
525 * Clean-up and extension of improvements for forked mode:
526 * - Child's return code now consists of flags RC_FLAG_*
527 * - Reporting toggle to parent now properly #ifdef'ed
528 * - Children now report blocking to parent. This enables
529 * statistics in forked mode
531 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
532 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
533 * Children now report having been toggled through _exit(17), parents
534 * watch for that code and toggle themselves if found.
536 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
537 * Wrapping all *_r functions (the non-_r versions of them) with
538 * mutex semaphores for OSX. Hopefully this will take care of all
539 * of those pesky crash reports.
541 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
542 * Fixed two mostly harmless mem leaks
544 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
545 * Fix: Head of global clients list now initialized to NULL
547 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
548 * Added basic support for OPTIONS and TRACE HTTP methods:
549 * - New interceptor direct_response() added in chat().
550 * - sed() moved to earlier in the process, so that the
551 * Host: header is evaluated before actions and forwarding
554 * Revision 1.92 2002/05/08 16:00:46 oes
555 * Chat's buffer handling:
556 * - Fixed bug with unchecked out-of-mem conditions
557 * while reading client request & server headers
558 * - No longer predict if the buffer limit will be exceeded
559 * in the next read -- check add_to_iob's new
560 * return code. If buffer couldn't be extended
561 * (policy or out-of-mem) while
562 * - reading from client: abort
563 * - reading server headers: send error page
564 * - buffering server body for filter: flush,
565 * and if that fails: send error page
567 * Revision 1.91 2002/04/08 20:35:58 swa
570 * Revision 1.90 2002/04/02 14:57:28 oes
571 * Made sending wafers independent of FEATURE_COOKIE_JAR
573 * Revision 1.89 2002/03/31 17:18:59 jongfoster
574 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
576 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
577 * More compiler warning message maintenance
579 * Revision 1.87 2002/03/26 22:29:54 swa
580 * we have a new homepage!
582 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
583 * Workaround for closing the jarfile before load_config() comes around again
585 * Revision 1.85 2002/03/24 15:23:33 jongfoster
588 * Revision 1.84 2002/03/24 13:25:43 swa
589 * name change related issues
591 * Revision 1.83 2002/03/16 23:54:06 jongfoster
592 * Adding graceful termination feature, to help look for memory leaks.
593 * If you enable this (which, by design, has to be done by hand
594 * editing config.h) and then go to http://i.j.b/die, then the program
595 * will exit cleanly after the *next* request. It should free all the
596 * memory that was used.
598 * Revision 1.82 2002/03/13 00:27:05 jongfoster
601 * Revision 1.81 2002/03/12 01:42:50 oes
602 * Introduced modular filters
604 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
605 * OS/2 port maintenance:
606 * - Fixed EMX build - it had decayed a little
607 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
608 * substituted a memset for now.
610 * Revision 1.79 2002/03/09 20:03:52 jongfoster
611 * - Making various functions return int rather than size_t.
612 * (Undoing a recent change). Since size_t is unsigned on
613 * Windows, functions like read_socket that return -1 on
614 * error cannot return a size_t.
616 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
617 * crashes, and also frequently caused JB to jump to 100%
618 * CPU and stay there. (Because it thought it had just
619 * read ((unsigned)-1) == 4Gb of data...)
621 * - The signature of write_socket has changed, it now simply
622 * returns success=0/failure=nonzero.
624 * - Trying to get rid of a few warnings --with-debug on
625 * Windows, I've introduced a new type "jb_socket". This is
626 * used for the socket file descriptors. On Windows, this
627 * is SOCKET (a typedef for unsigned). Everywhere else, it's
628 * an int. The error value can't be -1 any more, so it's
629 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
630 * Windows it maps to the #define INVALID_SOCKET.)
632 * - The signature of bind_port has changed.
634 * Revision 1.78 2002/03/08 21:35:04 oes
635 * Added optional group supplement to --user option. Will now use default group of user if no group given
637 * Revision 1.77 2002/03/07 03:52:06 oes
638 * - Fixed compiler warnings etc
639 * - Improved handling of failed DNS lookups
641 * Revision 1.76 2002/03/06 22:54:35 jongfoster
642 * Automated function-comment nitpicking.
644 * Revision 1.75 2002/03/06 10:02:19 oes
645 * Fixed stupid bug when --user was not given
647 * Revision 1.74 2002/03/06 00:49:31 jongfoster
648 * Fixing warning on Windows
649 * Making #ifdefs that refer to the same variable consistently
650 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
652 * Revision 1.73 2002/03/05 23:57:30 hal9
653 * Stray character 's' on line 1618 was breaking build.
655 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
656 * - Re-enable OS/2 building after new parms were added
657 * - Fix false out of memory report when resolving CGI templates when no IP
658 * address is available of failed attempt (a la no such domain)
660 * Revision 1.71 2002/03/05 18:13:56 oes
661 * Added --user option
663 * Revision 1.70 2002/03/05 04:52:42 oes
664 * Deleted non-errlog debugging code
666 * Revision 1.69 2002/03/04 23:50:00 jongfoster
667 * Splitting off bind_port() call into bind_port_helper(), with
670 * Revision 1.68 2002/03/04 20:17:32 oes
673 * Revision 1.67 2002/03/04 18:18:57 oes
674 * - Removed _DEBUG mode
675 * - Cleand up cmdline parsing
676 * - Introduced --no-daemon, --pidfile options
677 * - Cleaned up signal handling:
678 * - Terminate cleanly on INT, TERM and ABRT
679 * - Schedule logfile for re-opening on HUP
680 * - Ignore CHLD and PIPE
681 * - Leave the rest with their default handlers
682 * - Uniform handler registration
683 * - Added usage() function
684 * - Played styleguide police
686 * Revision 1.66 2002/03/03 15:06:55 oes
687 * Re-enabled automatic config reloading
689 * Revision 1.65 2002/03/03 14:49:11 oes
690 * Fixed CLF logging: Now uses client's original HTTP request
692 * Revision 1.64 2002/03/03 09:18:03 joergs
693 * Made jumbjuster work on AmigaOS again.
695 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
696 * Clean up a little CRLF unpleasantness that suddenly appeared
698 * Revision 1.62 2002/02/20 23:17:23 jongfoster
699 * Detecting some out-of memory conditions and exiting with a log message.
701 * Revision 1.61 2002/01/17 21:01:52 jongfoster
702 * Moving all our URL and URL pattern parsing code to urlmatch.c.
704 * Revision 1.60 2001/12/30 14:07:32 steudten
705 * - Add signal handling (unix)
706 * - Add SIGHUP handler (unix)
707 * - Add creation of pidfile (unix)
708 * - Add action 'top' in rc file (RH)
709 * - Add entry 'SIGNALS' to manpage
710 * - Add exit message to logfile (unix)
712 * Revision 1.59 2001/12/13 14:07:18 oes
713 * Fixed Bug: 503 error page now sent OK
715 * Revision 1.58 2001/11/30 23:37:24 jongfoster
716 * Renaming the Win32 config file to config.txt - this is almost the
717 * same as the corresponding UNIX name "config"
719 * Revision 1.57 2001/11/16 00:47:43 jongfoster
720 * Changing the tty-disconnection code to use setsid().
722 * Revision 1.56 2001/11/13 20:20:54 jongfoster
723 * Tabs->spaces, fixing a bug with missing {} around an if()
725 * Revision 1.55 2001/11/13 20:14:53 jongfoster
726 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
728 * Revision 1.54 2001/11/07 00:03:14 steudten
729 * Give reliable return value if an error
730 * occurs not just 0 with new daemon mode.
732 * Revision 1.53 2001/11/05 21:41:43 steudten
733 * Add changes to be a real daemon just for unix os.
734 * (change cwd to /, detach from controlling tty, set
735 * process group and session leader to the own process.
737 * Add some fatal-error log message for failed malloc().
738 * Add '-d' if compiled with 'configure --with-debug' to
739 * enable debug output.
741 * Revision 1.52 2001/10/26 20:11:20 jongfoster
742 * Fixing type mismatch
744 * Revision 1.51 2001/10/26 17:38:28 oes
747 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
748 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
749 * threads to call select() simultaneously. So, it's time to do a real, live,
750 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
751 * (native). Both versions will work, but using __OS2__ offers multi-threading.
753 * Revision 1.49 2001/10/23 21:41:35 jongfoster
754 * Added call to initialize the (statically-allocated of course)
755 * "out of memory" CGI response.
757 * Revision 1.48 2001/10/10 19:56:46 jongfoster
758 * Moving some code that wasn't cookie-related out of an #ifdef
761 * Revision 1.47 2001/10/10 16:44:36 oes
762 * Added CONNECT destination port limitation check
764 * Revision 1.46 2001/10/08 15:17:41 oes
765 * Re-enabled SSL forwarding
767 * Revision 1.45 2001/10/07 15:42:11 oes
768 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
770 * Moved downgrading of the HTTP version from parse_http_request to
771 * chat(), since we can't decide if it is necessary before we have
772 * determined the actions for the URL. The HTTP command is now
773 * *always* re-built so the repairs need no longer be special-cased.
775 * filter_popups now gets a csp pointer so it can raise the new
776 * CSP_FLAG_MODIFIED flag.
780 * Added configurable size limit for the IOB. If the IOB grows so
781 * large that the next read would exceed the limit, the header
782 * is generated, and the header & unfiltered buffer are flushed
783 * to the client. Chat then continues in non-buffering,
784 * non-filtering body mode.
786 * Revision 1.44 2001/10/02 18:13:57 oes
789 * Revision 1.43 2001/10/02 15:32:13 oes
790 * Moved generation of hdr
792 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
793 * Cleaning up 2 compiler warnings on OS/2.
795 * Revision 1.41 2001/09/16 17:05:14 jongfoster
796 * Removing unused #include showarg.h
798 * Revision 1.40 2001/09/16 15:41:45 jongfoster
799 * Fixing signed/unsigned comparison warning.
801 * Revision 1.39 2001/09/16 13:21:27 jongfoster
802 * Changes to use new list functions.
804 * Revision 1.38 2001/09/16 13:01:46 jongfoster
805 * Removing redundant function call that zeroed zalloc()'d memory.
807 * Revision 1.37 2001/09/10 11:12:24 oes
808 * Deleted unused variable
810 * Revision 1.36 2001/09/10 10:56:15 oes
811 * Silenced compiler warnings
813 * Revision 1.35 2001/07/31 14:44:22 oes
814 * Deleted unused size parameter from filter_popups()
816 * Revision 1.34 2001/07/30 22:08:36 jongfoster
817 * Tidying up #defines:
818 * - All feature #defines are now of the form FEATURE_xxx
819 * - Permanently turned off WIN_GUI_EDIT
820 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
822 * Revision 1.33 2001/07/29 19:32:00 jongfoster
823 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
825 * Revision 1.32 2001/07/29 18:47:05 jongfoster
826 * Adding missing #include "loadcfg.h"
828 * Revision 1.31 2001/07/29 12:17:48 oes
829 * Applied pthread fix by Paul Lieverse
831 * Revision 1.30 2001/07/25 22:57:13 jongfoster
832 * __BEOS__ no longer overrides FEATURE_PTHREAD.
833 * This is because FEATURE_PTHREAD will soon be widely used, so I
834 * want to keep it simple.
836 * Revision 1.29 2001/07/24 12:47:06 oes
837 * Applied BeOS support update by Eugenia
839 * Revision 1.28 2001/07/23 13:26:12 oes
840 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
842 * Revision 1.27 2001/07/19 19:09:47 haroon
843 * - Added code to take care of the situation where while processing the first
844 * server response (which includes the server header), after finding the end
845 * of the headers we were not looking past the end of the headers for
846 * content modification. I enabled it for filter_popups.
847 * Someone else should look to see if other similar operations should be
848 * done to the discarded portion of the buffer.
850 * Note 2001/07/20: No, the other content modification mechanisms will process
851 * the whole iob later anyway. --oes
853 * Revision 1.26 2001/07/18 12:31:36 oes
856 * Revision 1.25 2001/07/15 19:43:49 jongfoster
857 * Supports POSIX threads.
858 * Also removed some unused #includes.
860 * Revision 1.24 2001/07/13 14:00:40 oes
861 * - Generic content modification scheme:
862 * Each feature has its own applicability flag that is set
863 * from csp->action->flags.
864 * Replaced the "filtering" int flag , by a function pointer
865 * "content_filter" to the function that will do the content
866 * modification. If it is != NULL, the document will be buffered
867 * and processed through *content_filter, which must set
868 * csp->content_length and return a modified copy of the body
869 * or return NULL (on failiure).
870 * - Changed csp->is_text to the more generic bitmap csp->content_type
871 * which can currently take the valued CT_TEXT or CT_GIF
873 * - Removed all #ifdef PCRS
875 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
876 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
877 * the `block_acl' call. This prevents a compilation error when the user
878 * does not wish to use the "ACL" feature.
880 * Revision 1.22 2001/06/29 21:45:41 oes
881 * Indentation, CRLF->LF, Tab-> Space
883 * Revision 1.21 2001/06/29 13:29:36 oes
884 * - Cleaned up, improved comments
885 * - Unified all possible interceptors (CGI,
886 * block, trust, fast_redirect) in one
887 * place, with one (CGI) answer generation
888 * mechansim. Much clearer now.
889 * - Removed the GIF image generation, which
890 * is now done in filters.c:block_url()
891 * - Made error conditions like domain lookup
892 * failiure or (various) problems while talking
893 * to the server use cgi.c:error_response()
894 * instead of generating HTML/HTTP in chat() (yuck!)
895 * - Removed logentry from cancelled commit
897 * Revision 1.20 2001/06/09 10:55:28 jongfoster
898 * Changing BUFSIZ ==> BUFFER_SIZE
900 * Revision 1.19 2001/06/07 23:12:52 jongfoster
901 * Replacing function pointer in struct gateway with a directly
902 * called function forwarded_connect().
903 * Replacing struct gateway with struct forward_spec
905 * Revision 1.18 2001/06/03 19:12:16 oes
906 * introduced new cgi handling
908 * Revision 1.17 2001/06/01 20:07:23 jongfoster
909 * Now uses action +image-blocker{} rather than config->tinygif
911 * Revision 1.16 2001/06/01 18:49:17 jongfoster
912 * Replaced "list_share" with "list" - the tiny memory gain was not
913 * worth the extra complexity.
915 * Revision 1.15 2001/05/31 21:24:47 jongfoster
916 * Changed "permission" to "action" throughout.
917 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
918 * Moved vanilla wafer check into chat(), since we must now
919 * decide whether or not to add it based on the URL.
921 * Revision 1.14 2001/05/29 20:14:01 joergs
922 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
925 * Revision 1.13 2001/05/29 09:50:24 jongfoster
926 * Unified blocklist/imagelist/permissionslist.
927 * File format is still under discussion, but the internal changes
930 * Also modified interceptor behaviour:
931 * - We now intercept all URLs beginning with one of the following
932 * prefixes (and *only* these prefixes):
934 * * http://ijbswa.sf.net/config/
935 * * http://ijbswa.sourceforge.net/config/
936 * - New interceptors "home page" - go to http://i.j.b/ to see it.
937 * - Internal changes so that intercepted and fast redirect pages
938 * are not replaced with an image.
939 * - Interceptors now have the option to send a binary page direct
940 * to the client. (i.e. ijb-send-banner uses this)
941 * - Implemented show-url-info interceptor. (Which is why I needed
942 * the above interceptors changes - a typical URL is
943 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
944 * The previous mechanism would not have intercepted that, and
945 * if it had been intercepted then it then it would have replaced
948 * Revision 1.12 2001/05/27 22:17:04 oes
950 * - re_process_buffer no longer writes the modified buffer
951 * to the client, which was very ugly. It now returns the
952 * buffer, which it is then written by chat.
954 * - content_length now adjusts the Content-Length: header
955 * for modified documents rather than crunch()ing it.
956 * (Length info in csp->content_length, which is 0 for
957 * unmodified documents)
959 * - For this to work, sed() is called twice when filtering.
961 * Revision 1.11 2001/05/26 17:27:53 jongfoster
962 * Added support for CLF and fixed LOG_LEVEL_LOG.
963 * Also did CRLF->LF fix of my previous patch.
965 * Revision 1.10 2001/05/26 15:26:15 jongfoster
966 * ACL feature now provides more security by immediately dropping
967 * connections from untrusted hosts.
969 * Revision 1.9 2001/05/26 00:28:36 jongfoster
970 * Automatic reloading of config file.
971 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
972 * Most of the global variables have been moved to a new
973 * struct configuration_spec, accessed through csp->config->globalname
974 * Most of the globals remaining are used by the Win32 GUI.
976 * Revision 1.8 2001/05/25 22:43:18 jongfoster
977 * Fixing minor memory leak and buffer overflow.
979 * Revision 1.7 2001/05/25 22:34:30 jongfoster
982 * Revision 1.6 2001/05/23 00:13:58 joergs
983 * AmigaOS support fixed.
985 * Revision 1.5 2001/05/22 18:46:04 oes
987 * - Enabled filtering banners by size rather than URL
988 * by adding patterns that replace all standard banner
989 * sizes with the "Junkbuster" gif to the re_filterfile
991 * - Enabled filtering WebBugs by providing a pattern
992 * which kills all 1x1 images
994 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
995 * which is selected by the (nonstandard and therefore
996 * capital) letter 'U' in the option string.
997 * It causes the quantifiers to be ungreedy by default.
998 * Appending a ? turns back to greedy (!).
1000 * - Added a new interceptor ijb-send-banner, which
1001 * sends back the "Junkbuster" gif. Without imagelist or
1002 * MSIE detection support, or if tinygif = 1, or the
1003 * URL isn't recognized as an imageurl, a lame HTML
1004 * explanation is sent instead.
1006 * - Added new feature, which permits blocking remote
1007 * script redirects and firing back a local redirect
1009 * The feature is conditionally compiled, i.e. it
1010 * can be disabled with --disable-fast-redirects,
1011 * plus it must be activated by a "fast-redirects"
1012 * line in the config file, has its own log level
1013 * and of course wants to be displayed by show-proxy-args
1014 * Note: Boy, all the #ifdefs in 1001 locations and
1015 * all the fumbling with configure.in and acconfig.h
1016 * were *way* more work than the feature itself :-(
1018 * - Because a generic redirect template was needed for
1019 * this, tinygif = 3 now uses the same.
1021 * - Moved GIFs, and other static HTTP response templates
1024 * - Some minor fixes
1026 * - Removed some >400 CRs again (Jon, you really worked
1029 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1030 * Made failure to bind() a fatal error.
1032 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1033 * Version 2.9.4 checkin.
1034 * - Merged popupfile and cookiefile, and added control over PCRS
1035 * filtering, in new "permissionsfile".
1036 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1037 * file error you now get a message box (in the Win32 GUI) rather
1038 * than the program exiting with no explanation.
1039 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1041 * - Removed tabs from "config"
1042 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1043 * - Bumped up version number.
1045 * Revision 1.2 2001/05/17 22:34:44 oes
1046 * - Added hint on GIF char array generation to jcc.c
1047 * - Cleaned CRLF's from the sources and related files
1048 * - Repaired logging for REF and FRC
1050 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1051 * Initial import of version 2.9.3 source tree
1054 *********************************************************************/
1060 #include <sys/types.h>
1069 # ifndef FEATURE_PTHREAD
1073 # include <windows.h>
1074 # include <process.h>
1075 # endif /* ndef FEATURE_PTHREAD */
1078 # ifndef _WIN_CONSOLE
1079 # include "w32log.h"
1080 # endif /* ndef _WIN_CONSOLE */
1081 # include "w32svrapi.h"
1083 #else /* ifndef _WIN32 */
1085 # if !defined (__OS2__)
1086 # include <unistd.h>
1087 # include <sys/wait.h>
1088 # endif /* ndef __OS2__ */
1089 # include <sys/time.h>
1090 # include <sys/stat.h>
1091 # include <sys/ioctl.h>
1094 #include <sys/termios.h>
1102 # include <signal.h>
1105 # include <socket.h> /* BeOS has select() for sockets only. */
1106 # include <OS.h> /* declarations for threads and stuff. */
1109 # if defined(__EMX__) || defined(__OS2__)
1110 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1115 #define bzero(B,N) memset(B,0x00,n)
1119 # include <select.h>
1124 #include "project.h"
1127 #include "filters.h"
1128 #include "loaders.h"
1129 #include "parsers.h"
1130 #include "miscutil.h"
1132 #include "jbsockets.h"
1133 #include "gateway.h"
1134 #include "actions.h"
1136 #include "loadcfg.h"
1137 #include "urlmatch.h"
1139 const char jcc_h_rcs[] = JCC_H_VERSION;
1140 const char project_h_rcs[] = PROJECT_H_VERSION;
1143 struct client_state clients[1];
1144 struct file_list files[1];
1146 #ifdef FEATURE_STATISTICS
1147 int urls_read = 0; /* total nr of urls read inc rejected */
1148 int urls_rejected = 0; /* total nr of urls rejected */
1149 #endif /* def FEATURE_STATISTICS */
1151 #ifdef FEATURE_GRACEFUL_TERMINATION
1152 int g_terminate = 0;
1155 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1156 static void sig_handler(int the_signal);
1158 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1159 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1160 static jb_err get_server_headers(struct client_state *csp);
1161 static const char *crunch_reason(const struct http_response *rsp);
1162 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1164 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1166 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1167 static jb_err change_request_destination(struct client_state *csp);
1168 static void chat(struct client_state *csp);
1169 static void serve(struct client_state *csp);
1170 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1171 static void usage(const char *myname);
1173 static void initialize_mutexes(void);
1174 static jb_socket bind_port_helper(struct configuration_spec *config);
1175 static void listen_loop(void);
1178 void serve(struct client_state *csp);
1179 #else /* ifndef AMIGA */
1180 static void serve(struct client_state *csp);
1181 #endif /* def AMIGA */
1184 static int32 server_thread(void *data);
1185 #endif /* def __BEOS__ */
1188 #define sleep(N) Sleep(((N) * 1000))
1192 #define sleep(N) DosSleep(((N) * 100))
1195 #ifdef MUTEX_LOCKS_AVAILABLE
1197 * XXX: Does the locking stuff really belong in this file?
1199 privoxy_mutex_t log_mutex;
1200 privoxy_mutex_t log_init_mutex;
1201 privoxy_mutex_t connection_reuse_mutex;
1203 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1204 privoxy_mutex_t resolver_mutex;
1205 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1207 #ifndef HAVE_GMTIME_R
1208 privoxy_mutex_t gmtime_mutex;
1209 #endif /* ndef HAVE_GMTIME_R */
1211 #ifndef HAVE_LOCALTIME_R
1212 privoxy_mutex_t localtime_mutex;
1213 #endif /* ndef HAVE_GMTIME_R */
1216 privoxy_mutex_t rand_mutex;
1217 #endif /* ndef HAVE_RANDOM */
1219 #endif /* def MUTEX_LOCKS_AVAILABLE */
1222 const char *basedir = NULL;
1223 const char *pidfile = NULL;
1224 static int received_hup_signal = 0;
1225 #endif /* defined unix */
1227 /* HTTP snipplets. */
1228 static const char CSUCCEED[] =
1229 "HTTP/1.0 200 Connection established\r\n"
1230 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1232 static const char CHEADER[] =
1233 "HTTP/1.0 400 Invalid header received from client\r\n"
1234 "Proxy-Agent: Privoxy " VERSION "\r\n"
1235 "Content-Type: text/plain\r\n"
1236 "Connection: close\r\n\r\n"
1237 "Invalid header received from client.\r\n";
1239 static const char FTP_RESPONSE[] =
1240 "HTTP/1.0 400 Invalid request received from client\r\n"
1241 "Content-Type: text/plain\r\n"
1242 "Connection: close\r\n\r\n"
1243 "Invalid request. Privoxy doesn't support FTP.\r\n";
1245 static const char GOPHER_RESPONSE[] =
1246 "HTTP/1.0 400 Invalid request received from client\r\n"
1247 "Content-Type: text/plain\r\n"
1248 "Connection: close\r\n\r\n"
1249 "Invalid request. Privoxy doesn't support gopher.\r\n";
1251 /* XXX: should be a template */
1252 static const char MISSING_DESTINATION_RESPONSE[] =
1253 "HTTP/1.0 400 Bad request received from client\r\n"
1254 "Proxy-Agent: Privoxy " VERSION "\r\n"
1255 "Content-Type: text/plain\r\n"
1256 "Connection: close\r\n\r\n"
1257 "Bad request. Privoxy was unable to extract the destination.\r\n";
1259 /* XXX: should be a template */
1260 static const char NO_SERVER_DATA_RESPONSE[] =
1261 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1262 "Proxy-Agent: Privoxy " VERSION "\r\n"
1263 "Content-Type: text/plain\r\n"
1264 "Connection: close\r\n\r\n"
1265 "Empty server or forwarder response.\r\n"
1266 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1268 /* XXX: should be a template */
1269 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1270 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1271 "Proxy-Agent: Privoxy " VERSION "\r\n"
1272 "Content-Type: text/plain\r\n"
1273 "Connection: close\r\n\r\n"
1274 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1277 /* XXX: should be a template */
1278 static const char NULL_BYTE_RESPONSE[] =
1279 "HTTP/1.0 400 Bad request received from client\r\n"
1280 "Proxy-Agent: Privoxy " VERSION "\r\n"
1281 "Content-Type: text/plain\r\n"
1282 "Connection: close\r\n\r\n"
1283 "Bad request. Null byte(s) before end of request.\r\n";
1286 /* XXX: should be a template */
1287 static const char MESSED_UP_REQUEST_RESPONSE[] =
1288 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1289 "Proxy-Agent: Privoxy " VERSION "\r\n"
1290 "Content-Type: text/plain\r\n"
1291 "Connection: close\r\n\r\n"
1292 "Bad request. Messed up with header filters.\r\n";
1294 /* A function to crunch a response */
1295 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1297 /* Crunch function flags */
1298 #define CF_NO_FLAGS 0
1299 /* Cruncher applies to forced requests as well */
1300 #define CF_IGNORE_FORCE 1
1301 /* Crunched requests are counted for the block statistics */
1302 #define CF_COUNT_AS_REJECT 2
1304 /* A crunch function and its flags */
1307 const crunch_func_ptr cruncher;
1311 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1313 /* Complete list of cruncher functions */
1314 static const struct cruncher crunchers_all[] = {
1315 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1316 { block_url, CF_COUNT_AS_REJECT },
1317 #ifdef FEATURE_TRUST
1318 { trust_url, CF_COUNT_AS_REJECT },
1319 #endif /* def FEATURE_TRUST */
1320 { redirect_url, CF_NO_FLAGS },
1321 { dispatch_cgi, CF_IGNORE_FORCE},
1325 /* Light version, used after tags are applied */
1326 static const struct cruncher crunchers_light[] = {
1327 { block_url, CF_COUNT_AS_REJECT },
1328 { redirect_url, CF_NO_FLAGS },
1334 * XXX: Don't we really mean
1340 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1341 /*********************************************************************
1343 * Function : sig_handler
1345 * Description : Signal handler for different signals.
1346 * Exit gracefully on TERM and INT
1347 * or set a flag that will cause the errlog
1348 * to be reopened by the main thread on HUP.
1351 * 1 : the_signal = the signal cause this function to call
1355 *********************************************************************/
1356 static void sig_handler(int the_signal)
1362 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1374 received_hup_signal = 1;
1380 * We shouldn't be here, unless we catch signals
1381 * in main() that we can't handle here!
1383 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1391 /*********************************************************************
1393 * Function : client_protocol_is_unsupported
1395 * Description : Checks if the client used a known unsupported
1396 * protocol and deals with it by sending an error
1400 * 1 : csp = Current client state (buffers, headers, etc...)
1401 * 2 : req = the first request line send by the client
1403 * Returns : TRUE if an error response has been generated, or
1404 * FALSE if the request doesn't look invalid.
1406 *********************************************************************/
1407 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1410 * If it's a FTP or gopher request, we don't support it.
1412 * These checks are better than nothing, but they might
1413 * not work in all configurations and some clients might
1414 * have problems digesting the answer.
1416 * They should, however, never cause more problems than
1417 * Privoxy's old behaviour (returning the misleading HTML
1420 * "Could not resolve http://(ftp|gopher)://example.org").
1422 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1424 const char *response = NULL;
1425 const char *protocol = NULL;
1427 if (!strncmpic(req, "GET ftp://", 10))
1429 response = FTP_RESPONSE;
1434 response = GOPHER_RESPONSE;
1435 protocol = "GOPHER";
1437 log_error(LOG_LEVEL_ERROR,
1438 "%s tried to use Privoxy as %s proxy: %s",
1439 csp->ip_addr_str, protocol, req);
1440 log_error(LOG_LEVEL_CLF,
1441 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1443 write_socket(csp->cfd, response, strlen(response));
1452 /*********************************************************************
1454 * Function : get_request_destination_elsewhere
1456 * Description : If the client's request was redirected into
1457 * Privoxy without the client's knowledge,
1458 * the request line lacks the destination host.
1460 * This function tries to get it elsewhere,
1461 * provided accept-intercepted-requests is enabled.
1463 * "Elsewhere" currently only means "Host: header",
1464 * but in the future we may ask the redirecting
1465 * packet filter to look the destination up.
1467 * If the destination stays unknown, an error
1468 * response is send to the client and headers
1469 * are freed so that chat() can return directly.
1472 * 1 : csp = Current client state (buffers, headers, etc...)
1473 * 2 : headers = a header list
1475 * Returns : JB_ERR_OK if the destination is now known, or
1476 * JB_ERR_PARSE if it isn't.
1478 *********************************************************************/
1479 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1483 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1485 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1486 " Privoxy isn't configured to accept intercepted requests.",
1487 csp->ip_addr_str, csp->http->cmd);
1488 /* XXX: Use correct size */
1489 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1490 csp->ip_addr_str, csp->http->cmd);
1492 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1493 destroy_list(headers);
1495 return JB_ERR_PARSE;
1497 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1499 /* Split the domain we just got for pattern matching */
1500 init_domain_components(csp->http);
1506 /* We can't work without destination. Go spread the news.*/
1508 req = list_to_text(headers);
1510 /* XXX: Use correct size */
1511 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1512 csp->ip_addr_str, csp->http->cmd);
1513 log_error(LOG_LEVEL_ERROR,
1514 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1515 csp->ip_addr_str, csp->http->cmd, req);
1518 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1519 destroy_list(headers);
1521 return JB_ERR_PARSE;
1524 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1525 * to get the destination IP address, use it as host directly
1526 * or do a reverse DNS lookup first.
1531 /*********************************************************************
1533 * Function : get_server_headers
1535 * Description : Parses server headers in iob and fills them
1536 * into csp->headers so that they can later be
1540 * 1 : csp = Current client state (buffers, headers, etc...)
1542 * Returns : JB_ERR_OK if everything went fine, or
1543 * JB_ERR_PARSE if the headers were incomplete.
1545 *********************************************************************/
1546 static jb_err get_server_headers(struct client_state *csp)
1548 int continue_hack_in_da_house = 0;
1551 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1556 * continue hack in da house. Ignore the ending of
1557 * this head and continue enlisting header lines.
1558 * The reason is described below.
1560 enlist(csp->headers, "");
1561 continue_hack_in_da_house = 0;
1564 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1567 * It's a bodyless continue response, don't
1568 * stop header parsing after reaching its end.
1570 * As a result Privoxy will concatenate the
1571 * next response's head and parse and deliver
1572 * the headers as if they belonged to one request.
1574 * The client will separate them because of the
1575 * empty line between them.
1577 * XXX: What we're doing here is clearly against
1578 * the intended purpose of the continue header,
1579 * and under some conditions (HTTP/1.0 client request)
1580 * it's a standard violation.
1582 * Anyway, "sort of against the spec" is preferable
1583 * to "always getting confused by Continue responses"
1584 * (Privoxy's behaviour before this hack was added)
1586 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1587 continue_hack_in_da_house = 1;
1589 else if (*header == '\0')
1592 * If the header is empty, but the Continue hack
1593 * isn't active, we can assume that we reached the
1594 * end of the buffer before we hit the end of the
1597 * Inform the caller an let it decide how to handle it.
1599 return JB_ERR_PARSE;
1602 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1605 * XXX: Should we quit the request and return a
1606 * out of memory error page instead?
1608 log_error(LOG_LEVEL_ERROR,
1609 "Out of memory while enlisting server headers. %s lost.",
1619 /*********************************************************************
1621 * Function : crunch_reason
1623 * Description : Translates the crunch reason code into a string.
1626 * 1 : rsp = a http_response
1628 * Returns : A string with the crunch reason or an error description.
1630 *********************************************************************/
1631 static const char *crunch_reason(const struct http_response *rsp)
1633 char * reason = NULL;
1635 assert(rsp != NULL);
1638 return "Internal error while searching for crunch reason";
1641 switch (rsp->reason)
1643 case RSP_REASON_UNSUPPORTED:
1644 reason = "Unsupported HTTP feature";
1646 case RSP_REASON_BLOCKED:
1649 case RSP_REASON_UNTRUSTED:
1650 reason = "Untrusted";
1652 case RSP_REASON_REDIRECTED:
1653 reason = "Redirected";
1655 case RSP_REASON_CGI_CALL:
1656 reason = "CGI Call";
1658 case RSP_REASON_NO_SUCH_DOMAIN:
1659 reason = "DNS failure";
1661 case RSP_REASON_FORWARDING_FAILED:
1662 reason = "Forwarding failed";
1664 case RSP_REASON_CONNECT_FAILED:
1665 reason = "Connection failure";
1667 case RSP_REASON_OUT_OF_MEMORY:
1668 reason = "Out of memory (may mask other reasons)";
1671 reason = "No reason recorded";
1679 /*********************************************************************
1681 * Function : send_crunch_response
1683 * Description : Delivers already prepared response for
1684 * intercepted requests, logs the interception
1685 * and frees the response.
1688 * 1 : csp = Current client state (buffers, headers, etc...)
1689 * 1 : rsp = Fully prepared response. Will be freed on exit.
1691 * Returns : Nothing.
1693 *********************************************************************/
1694 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1696 const struct http_request *http = csp->http;
1697 char status_code[4];
1699 assert(rsp != NULL);
1700 assert(rsp->head != NULL);
1705 * Not supposed to happen. If it does
1706 * anyway, treat it as an unknown error.
1708 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1709 /* return code doesn't matter */
1714 /* If rsp is still NULL, we have serious internal problems. */
1715 log_error(LOG_LEVEL_FATAL,
1716 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1720 * Extract the status code from the actual head
1721 * that was send to the client. It is the only
1722 * way to get it right for all requests, including
1723 * the fixed ones for out-of-memory problems.
1725 * A head starts like this: 'HTTP/1.1 200...'
1729 status_code[0] = rsp->head[9];
1730 status_code[1] = rsp->head[10];
1731 status_code[2] = rsp->head[11];
1732 status_code[3] = '\0';
1734 /* Write the answer to the client */
1735 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1736 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1738 /* There is nothing we can do about it. */
1739 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1742 /* Log that the request was crunched and why. */
1743 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1744 http->hostport, http->path, crunch_reason(rsp));
1745 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1746 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1748 /* Clean up and return */
1749 if (cgi_error_memory() != rsp)
1751 free_http_response(rsp);
1758 /*********************************************************************
1760 * Function : request_contains_null_bytes
1762 * Description : Checks for NULL bytes in the request and sends
1763 * an error message to the client if any were found.
1765 * XXX: currently not used, see comment in chat().
1768 * 1 : csp = Current client state (buffers, headers, etc...)
1769 * 2 : buf = Data from the client's request to check.
1770 * 3 : len = The data length.
1772 * Returns : TRUE if the request contained one or more NULL bytes, or
1775 *********************************************************************/
1776 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1778 size_t c_len; /* Request lenght when treated as C string */
1780 c_len = strlen(buf);
1785 * Null byte(s) found. Log the request,
1786 * return an error response and hang up.
1788 size_t tmp_len = c_len;
1793 * Replace NULL byte(s) with '°' characters
1794 * so the request can be logged as string.
1795 * XXX: Is there a better replacement character?
1798 tmp_len += strlen(buf+tmp_len);
1799 } while (tmp_len < len);
1801 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1802 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1803 log_error(LOG_LEVEL_HEADER,
1804 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1806 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1808 /* XXX: Log correct size */
1809 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1819 /*********************************************************************
1821 * Function : crunch_response_triggered
1823 * Description : Checks if the request has to be crunched,
1824 * and delivers the crunch response if necessary.
1827 * 1 : csp = Current client state (buffers, headers, etc...)
1828 * 2 : crunchers = list of cruncher functions to run
1830 * Returns : TRUE if the request was answered with a crunch response
1833 *********************************************************************/
1834 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1836 struct http_response *rsp = NULL;
1837 const struct cruncher *c;
1840 * If CGI request crunching is disabled,
1841 * check the CGI dispatcher out of order to
1842 * prevent unintentional blocks or redirects.
1844 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1845 && (NULL != (rsp = dispatch_cgi(csp))))
1847 /* Deliver, log and free the interception response. */
1848 send_crunch_response(csp, rsp);
1852 for (c = crunchers; c->cruncher != NULL; c++)
1855 * Check the cruncher if either Privoxy is toggled
1856 * on and the request isn't forced, or if the cruncher
1857 * applies to forced requests as well.
1859 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1860 !(csp->flags & CSP_FLAG_FORCED)) ||
1861 (c->flags & CF_IGNORE_FORCE))
1863 rsp = c->cruncher(csp);
1866 /* Deliver, log and free the interception response. */
1867 send_crunch_response(csp, rsp);
1868 #ifdef FEATURE_STATISTICS
1869 if (c->flags & CF_COUNT_AS_REJECT)
1871 csp->flags |= CSP_FLAG_REJECTED;
1873 #endif /* def FEATURE_STATISTICS */
1884 /*********************************************************************
1886 * Function : build_request_line
1888 * Description : Builds the HTTP request line.
1890 * If a HTTP forwarder is used it expects the whole URL,
1891 * web servers only get the path.
1894 * 1 : csp = Current client state (buffers, headers, etc...)
1895 * 2 : fwd = The forwarding spec used for the request
1896 * XXX: Should use http->fwd instead.
1897 * 3 : request_line = The old request line which will be replaced.
1899 * Returns : Nothing. Terminates in case of memory problems.
1901 *********************************************************************/
1902 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1904 struct http_request *http = csp->http;
1906 assert(http->ssl == 0);
1909 * Downgrade http version from 1.1 to 1.0
1910 * if +downgrade action applies.
1912 if ( (csp->action->flags & ACTION_DOWNGRADE)
1913 && (!strcmpic(http->ver, "HTTP/1.1")))
1916 http->ver = strdup("HTTP/1.0");
1918 if (http->ver == NULL)
1920 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1925 * Rebuild the request line.
1927 freez(*request_line);
1928 *request_line = strdup(http->gpc);
1929 string_append(request_line, " ");
1931 if (fwd->forward_host)
1933 string_append(request_line, http->url);
1937 string_append(request_line, http->path);
1939 string_append(request_line, " ");
1940 string_append(request_line, http->ver);
1942 if (*request_line == NULL)
1944 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1946 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1950 /*********************************************************************
1952 * Function : change_request_destination
1954 * Description : Parse a (rewritten) request line and regenerate
1955 * the http request data.
1958 * 1 : csp = Current client state (buffers, headers, etc...)
1960 * Returns : Forwards the parse_http_request() return code.
1961 * Terminates in case of memory problems.
1963 *********************************************************************/
1964 static jb_err change_request_destination(struct client_state *csp)
1966 struct http_request *http = csp->http;
1969 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1970 free_http_request(http);
1971 err = parse_http_request(csp->headers->first->str, http, csp);
1972 if (JB_ERR_OK != err)
1974 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1975 jb_err_to_string(err));
1979 /* XXX: ocmd is a misleading name */
1980 http->ocmd = strdup(http->cmd);
1981 if (http->ocmd == NULL)
1983 log_error(LOG_LEVEL_FATAL,
1984 "Out of memory copying rewritten HTTP request line");
1992 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
1993 /*********************************************************************
1995 * Function : server_response_is_complete
1997 * Description : Determines whether we should stop reading
1998 * from the server socket.
2001 * 1 : csp = Current client state (buffers, headers, etc...)
2002 * 2 : content_length = Length of content received so far.
2004 * Returns : TRUE if the response is complete,
2007 *********************************************************************/
2008 static int server_response_is_complete(struct client_state *csp, size_t content_length)
2010 int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2012 if (!strcmpic(csp->http->gpc, "HEAD"))
2015 * "HEAD" implies no body, we are thus expecting
2016 * no content. XXX: incomplete "list" of methods?
2018 log_error(LOG_LEVEL_INFO, "Method %s implies no body.", csp->http->gpc);
2019 csp->expected_content_length = 0;
2020 content_length_known = TRUE;
2023 if (csp->http->status == 304)
2026 * Expect no body. XXX: incomplete "list" of status codes?
2028 log_error(LOG_LEVEL_INFO, "Status code %d implies no body.", csp->http->status);
2029 csp->expected_content_length = 0;
2030 content_length_known = TRUE;
2033 return (content_length_known && ((0 == csp->expected_content_length)
2034 || (csp->expected_content_length <= content_length)));
2036 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2039 /*********************************************************************
2043 * Description : Once a connection to the client has been accepted,
2044 * this function is called (via serve()) to handle the
2045 * main business of the communication. When this
2046 * function returns, the caller must close the client
2049 * FIXME: chat is nearly thousand lines long.
2053 * 1 : csp = Current client state (buffers, headers, etc...)
2055 * Returns : Nothing.
2057 *********************************************************************/
2058 static void chat(struct client_state *csp)
2060 char buf[BUFFER_SIZE];
2068 int ms_iis5_hack = 0;
2069 size_t byte_count = 0;
2070 int forwarded_connect_retries = 0;
2071 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2072 const struct forward_spec * fwd;
2073 struct http_request *http;
2074 int len; /* for buffer sizes (and negative error codes) */
2077 /* Function that does the content filtering for the current request */
2078 filter_function_ptr content_filter = NULL;
2080 /* Skeleton for HTTP response, if we should intercept the request */
2081 struct http_response *rsp;
2083 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2084 struct list header_list;
2085 struct list *headers = &header_list;
2089 memset(buf, 0, sizeof(buf));
2092 * Read the client's request. Note that since we're not using select() we
2093 * could get blocked here if a client connected, then didn't say anything!
2098 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2100 if (len <= 0) break; /* error! */
2103 * If there is no memory left for buffering the
2104 * request, there is nothing we can do but hang up
2106 if (add_to_iob(csp, buf, len))
2111 req = get_header(csp->iob);
2113 } while ((NULL != req) && ('\0' == *req));
2115 if ((NULL != req) && ('\0' != *req))
2117 /* Request received. Validate and parse it. */
2121 * XXX: Temporary disabled to prevent problems
2122 * with POST requests whose bodies are allowed to
2123 * contain NULL bytes. BR#1730105.
2125 * The main purpose of this check is to properly
2126 * log stuff like BitTorrent traffic and other junk
2127 * that hits public proxies. It's not required for
2128 * Privoxy to functions as those requests are discarded
2131 * It probably should be rewritten to only check
2132 * the head of the request. Another option would
2133 * be to let all POST requests pass, although that
2134 * may not be good enough.
2136 if (request_contains_null_bytes(csp, buf, len))
2138 /* NULL bytes found and dealt with, just hang up. */
2143 /* Does the request line look invalid? */
2144 if (client_protocol_is_unsupported(csp, req))
2147 * Yes. The request has already been
2148 * answered with a error response, the buffers
2149 * were freed and we're done with chatting.
2154 #ifdef FEATURE_FORCE_LOAD
2156 * If this request contains the FORCE_PREFIX and blocks
2157 * aren't enforced, get rid of it and set the force flag.
2159 if (strstr(req, FORCE_PREFIX))
2161 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2163 log_error(LOG_LEVEL_FORCE,
2164 "Ignored force prefix in request: \"%s\".", req);
2168 strclean(req, FORCE_PREFIX);
2169 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2170 csp->flags |= CSP_FLAG_FORCED;
2174 #endif /* def FEATURE_FORCE_LOAD */
2175 err = parse_http_request(req, http, csp);
2176 if (JB_ERR_OK != err)
2178 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2184 if (http->cmd == NULL)
2186 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2187 /* XXX: Use correct size */
2188 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2189 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2191 free_http_request(http);
2195 /* grab the rest of the client's headers */
2199 p = get_header(csp->iob);
2203 /* There are no additional headers to read. */
2210 * We didn't receive a complete header
2211 * line yet, get the rest of it.
2213 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2216 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2217 destroy_list(headers);
2221 if (add_to_iob(csp, buf, len))
2224 * If there is no memory left for buffering the
2225 * request, there is nothing we can do but hang up
2227 destroy_list(headers);
2234 * We were able to read a complete
2235 * header and can finaly enlist it.
2242 if (http->host == NULL)
2245 * If we still don't know the request destination,
2246 * the request is invalid or the client uses
2247 * Privoxy without its knowledge.
2249 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2252 * Our attempts to get the request destination
2253 * elsewhere failed or Privoxy is configured
2254 * to only accept proxy requests.
2256 * An error response has already been send
2257 * and we're done here.
2264 * Determine the actions for this URL
2266 #ifdef FEATURE_TOGGLE
2267 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2269 /* Most compatible set of actions (i.e. none) */
2270 init_current_action(csp->action);
2273 #endif /* ndef FEATURE_TOGGLE */
2275 get_url_actions(csp, http);
2279 * Save a copy of the original request for logging
2281 http->ocmd = strdup(http->cmd);
2283 if (http->ocmd == NULL)
2285 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2288 enlist(csp->headers, http->cmd);
2290 /* Append the previously read headers */
2291 list_append_list_unique(csp->headers, headers);
2292 destroy_list(headers);
2294 err = sed(csp, FILTER_CLIENT_HEADERS);
2295 if (JB_ERR_OK != err)
2297 assert(err == JB_ERR_PARSE);
2298 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2300 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2302 /* Check request line for rewrites. */
2303 if ((NULL == csp->headers->first->str)
2304 || (strcmp(http->cmd, csp->headers->first->str) &&
2305 (JB_ERR_OK != change_request_destination(csp))))
2308 * A header filter broke the request line - bail out.
2310 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2311 /* XXX: Use correct size */
2312 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2313 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2315 free_http_request(http);
2319 /* decide how to route the HTTP request */
2320 fwd = forward_url(csp, http);
2323 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2324 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2328 * build the http request to send to the server
2329 * we have to do one of the following:
2331 * create = use the original HTTP request to create a new
2332 * HTTP request that has either the path component
2333 * without the http://domainspec (w/path) or the
2334 * full orininal URL (w/url)
2335 * Note that the path and/or the HTTP version may
2336 * have been altered by now.
2338 * connect = Open a socket to the host:port of the server
2339 * and short-circuit server and client socket.
2341 * pass = Pass the request unchanged if forwarding a CONNECT
2342 * request to a parent proxy. Note that we'll be sending
2343 * the CFAIL message ourselves if connecting to the parent
2344 * fails, but we won't send a CSUCCEED message if it works,
2345 * since that would result in a double message (ours and the
2346 * parent's). After sending the request to the parent, we simply
2349 * here's the matrix:
2352 * +--------+--------+
2354 * 0 | create | connect|
2356 * Forwarding +--------+--------+
2358 * 1 | create | pass |
2360 * +--------+--------+
2364 if (http->ssl && connect_port_is_forbidden(csp))
2366 const char *acceptable_connect_ports =
2367 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2368 assert(NULL != acceptable_connect_ports);
2369 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2370 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2371 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2372 csp->action->flags |= ACTION_BLOCK;
2378 freez(csp->headers->first->str);
2379 build_request_line(csp, fwd, &csp->headers->first->str);
2382 hdr = list_to_text(csp->headers);
2385 /* FIXME Should handle error properly */
2386 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2390 * We have a request. Check if one of the crunchers wants it.
2392 if (crunch_response_triggered(csp, crunchers_all))
2395 * Yes. The client got the crunch response
2396 * and we are done here after cleaning up.
2399 list_remove_all(csp->headers);
2405 * The headers can't be removed earlier because
2406 * they were still needed for the referrer check
2407 * in case of CGI crunches.
2409 * XXX: Would it be worth to move the referrer check
2410 * into client_referrer() and set a flag if it's trusted?
2412 list_remove_all(csp->headers);
2414 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2416 if (fwd->forward_host)
2418 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2419 fwd->forward_host, fwd->forward_port, http->hostport);
2423 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2426 /* here we connect to the server, gateway, or the forwarder */
2428 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2429 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2431 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2432 forwarded_connect_retries, http->hostport);
2435 if (csp->sfd == JB_INVALID_SOCKET)
2437 if (fwd->type != SOCKS_NONE)
2440 rsp = error_response(csp, "forwarding-failed", errno);
2442 else if (errno == EINVAL)
2444 rsp = error_response(csp, "no-such-domain", errno);
2448 rsp = error_response(csp, "connect-failed", errno);
2449 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2453 /* Write the answer to the client */
2456 send_crunch_response(csp, rsp);
2463 if (fwd->forward_host || (http->ssl == 0))
2466 * Write the client's (modified) header to the server
2467 * (along with anything else that may be in the buffer)
2469 if (write_socket(csp->sfd, hdr, strlen(hdr))
2470 || (flush_socket(csp->sfd, csp->iob) < 0))
2472 log_error(LOG_LEVEL_CONNECT,
2473 "write header to: %s failed: %E", http->hostport);
2475 rsp = error_response(csp, "connect-failed", errno);
2478 send_crunch_response(csp, rsp);
2488 * We're running an SSL tunnel and we're not forwarding,
2489 * so just send the "connect succeeded" message to the
2490 * client, flush the rest, and get out of the way.
2492 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2500 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2502 /* we're finished with the client's header */
2505 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2507 /* pass data between the client and server
2508 * until one or the other shuts down the connection.
2517 * FD_ZERO here seems to point to an errant macro which crashes.
2518 * So do this by hand for now...
2520 memset(&rfds,0x00,sizeof(fd_set));
2524 FD_SET(csp->cfd, &rfds);
2525 FD_SET(csp->sfd, &rfds);
2527 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2528 if (server_body && server_response_is_complete(csp, byte_count))
2530 log_error(LOG_LEVEL_CONNECT,
2531 "Done reading from server. Expected content length: %d. "
2532 "Actual content length: %d. Most recently received: %d.",
2533 csp->expected_content_length, byte_count, len);
2536 * XXX: should not jump around,
2537 * chat() is complicated enough already.
2541 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2543 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2547 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2552 * This is the body of the browser's request,
2553 * just read and write it.
2555 if (FD_ISSET(csp->cfd, &rfds))
2557 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2561 break; /* "game over, man" */
2564 if (write_socket(csp->sfd, buf, (size_t)len))
2566 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2573 * The server wants to talk. It could be the header or the body.
2574 * If `hdr' is null, then it's the header otherwise it's the body.
2575 * FIXME: Does `hdr' really mean `host'? No.
2577 if (FD_ISSET(csp->sfd, &rfds))
2580 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2584 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2586 if (http->ssl && (fwd->forward_host == NULL))
2589 * Just hang up. We already confirmed the client's CONNECT
2590 * request with status code 200 and unencrypted content is
2591 * no longer welcome.
2593 log_error(LOG_LEVEL_ERROR,
2594 "CONNECT already confirmed. Unable to tell the client about the problem.");
2597 else if (byte_count)
2600 * Just hang up. We already transmitted the original headers
2601 * and parts of the original content and therefore missed the
2602 * chance to send an error message (without risking data corruption).
2604 * XXX: we could retry with a fancy range request here.
2606 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2607 "Unable to tell the client about the problem.");
2611 rsp = error_response(csp, "connect-failed", errno);
2614 send_crunch_response(csp, rsp);
2620 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2621 if (csp->flags & CSP_FLAG_CHUNKED)
2623 if ((len > 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
2625 /* XXX: this is a temporary hack */
2626 log_error(LOG_LEVEL_CONNECT,
2627 "Looks like we reached the end of the last chunk: "
2628 "%d %d %d %d %d. We better stop reading.",
2629 buf[len-5], buf[len-4], buf[len-3], buf[len-2], buf[len-1]);
2630 csp->expected_content_length = byte_count + len;
2631 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2635 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2638 * Add a trailing zero to let be able to use string operations.
2639 * XXX: do we still need this with filter_popups gone?
2644 * Normally, this would indicate that we've read
2645 * as much as the server has sent us and we can
2646 * close the client connection. However, Microsoft
2647 * in its wisdom has released IIS/5 with a bug that
2648 * prevents it from sending the trailing \r\n in
2649 * a 302 redirect header (and possibly other headers).
2650 * To work around this if we've haven't parsed
2651 * a full header we'll append a trailing \r\n
2652 * and see if this now generates a valid one.
2654 * This hack shouldn't have any impacts. If we've
2655 * already transmitted the header or if this is a
2656 * SSL connection, then we won't bother with this
2657 * hack. So we only work on partially received
2658 * headers. If we append a \r\n and this still
2659 * doesn't generate a valid header, then we won't
2660 * transmit anything to the client.
2665 if (server_body || http->ssl)
2668 * If we have been buffering up the document,
2669 * now is the time to apply content modification
2670 * and send the result to the client.
2674 p = execute_content_filter(csp, content_filter);
2676 * If the content filter fails, use the original
2677 * buffer and length.
2678 * (see p != NULL ? p : csp->iob->cur below)
2682 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2685 if (JB_ERR_OK != update_server_headers(csp))
2687 log_error(LOG_LEVEL_FATAL,
2688 "Failed to update server headers. after filtering.");
2691 hdr = list_to_text(csp->headers);
2694 /* FIXME Should handle error properly */
2695 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2698 if (write_socket(csp->cfd, hdr, strlen(hdr))
2699 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2701 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2711 break; /* "game over, man" */
2715 * This is NOT the body, so
2716 * Let's pretend the server just sent us a blank line.
2718 snprintf(buf, sizeof(buf), "\r\n");
2719 len = (int)strlen(buf);
2722 * Now, let the normal header parsing algorithm below do its
2723 * job. If it fails, we'll exit instead of continuing.
2730 * If this is an SSL connection or we're in the body
2731 * of the server document, just write it to the client,
2732 * unless we need to buffer the body for later content-filtering
2734 if (server_body || http->ssl)
2739 * If there is no memory left for buffering the content, or the buffer limit
2740 * has been reached, switch to non-filtering mode, i.e. make & write the
2741 * header, flush the iob and buf, and get out of the way.
2743 if (add_to_iob(csp, buf, len))
2748 log_error(LOG_LEVEL_INFO,
2749 "Flushing header and buffers. Stepping back from filtering.");
2751 hdr = list_to_text(csp->headers);
2755 * Memory is too tight to even generate the header.
2756 * Send our static "Out-of-memory" page.
2758 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2759 rsp = cgi_error_memory();
2760 send_crunch_response(csp, rsp);
2764 hdrlen = strlen(hdr);
2766 if (write_socket(csp->cfd, hdr, hdrlen)
2767 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2768 || (write_socket(csp->cfd, buf, (size_t)len)))
2770 log_error(LOG_LEVEL_CONNECT,
2771 "Flush header and buffers to client failed: %E");
2777 * Reset the byte_count to the amount of bytes
2778 * we just flushed. len will be added a few lines below,
2779 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2781 byte_count = (size_t)flushed;
2783 content_filter = NULL;
2789 if (write_socket(csp->cfd, buf, (size_t)len))
2791 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2795 byte_count += (size_t)len;
2801 * We're still looking for the end of the server's header.
2802 * Buffer up the data we just read. If that fails, there's
2803 * little we can do but send our static out-of-memory page.
2805 if (add_to_iob(csp, buf, len))
2807 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2808 rsp = cgi_error_memory();
2809 send_crunch_response(csp, rsp);
2814 /* Convert iob into something sed() can digest */
2815 if (JB_ERR_PARSE == get_server_headers(csp))
2820 * Well, we tried our MS IIS/5 hack and it didn't work.
2821 * The header is incomplete and there isn't anything
2822 * we can do about it.
2824 log_error(LOG_LEVEL_INFO,
2825 "MS IIS5 hack didn't produce valid headers.");
2831 * Since we have to wait for more from the server before
2832 * we can parse the headers we just continue here.
2838 /* Did we actually get anything? */
2839 if (NULL == csp->headers->first)
2841 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2842 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2843 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2844 free_http_request(http);
2848 assert(csp->headers->first->str);
2850 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2851 strncmpic(csp->headers->first->str, "ICY", 3))
2854 * It doesn't look like a HTTP (or Shoutcast) response:
2855 * tell the client and log the problem.
2857 if (strlen(csp->headers->first->str) > 30)
2859 csp->headers->first->str[30] = '\0';
2861 log_error(LOG_LEVEL_ERROR,
2862 "Invalid server or forwarder response. Starts with: %s",
2863 csp->headers->first->str);
2864 log_error(LOG_LEVEL_CLF,
2865 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2866 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2867 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2868 free_http_request(http);
2873 * We have now received the entire server header,
2874 * filter it and send the result to the client
2876 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2878 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2880 hdr = list_to_text(csp->headers);
2883 /* FIXME Should handle error properly */
2884 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2887 if (crunch_response_triggered(csp, crunchers_light))
2890 * One of the tags created by a server-header
2891 * tagger triggered a crunch. We already
2892 * delivered the crunch response to the client
2893 * and are done here after cleaning up.
2898 /* Buffer and pcrs filter this if appropriate. */
2900 if (!http->ssl) /* We talk plaintext */
2902 content_filter = get_filter_function(csp);
2905 * Only write if we're not buffering for content modification
2907 if (!content_filter)
2910 * Write the server's (modified) header to
2911 * the client (along with anything else that
2912 * may be in the buffer)
2915 if (write_socket(csp->cfd, hdr, strlen(hdr))
2916 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2918 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2921 * The write failed, so don't bother mentioning it
2922 * to the client... it probably can't hear us anyway.
2928 byte_count += (size_t)len;
2931 /* we're finished with the server's header */
2937 * If this was a MS IIS/5 hack then it means the server
2938 * has already closed the connection. Nothing more to read.
2943 log_error(LOG_LEVEL_INFO,
2944 "Closed server connection detected with MS IIS5 hack enabled.");
2951 return; /* huh? we should never get here */
2954 if (csp->content_length == 0)
2957 * If Privoxy didn't recalculate the Content-Lenght,
2958 * byte_count is still correct.
2960 csp->content_length = byte_count;
2963 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2964 csp->ip_addr_str, http->ocmd, csp->content_length);
2968 /*********************************************************************
2972 * Description : This is little more than chat. We only "serve" to
2973 * to close any socket that chat may have opened.
2976 * 1 : csp = Current client state (buffers, headers, etc...)
2980 *********************************************************************/
2982 void serve(struct client_state *csp)
2983 #else /* ifndef AMIGA */
2984 static void serve(struct client_state *csp)
2985 #endif /* def AMIGA */
2988 close_socket(csp->cfd);
2990 if (csp->sfd != JB_INVALID_SOCKET)
2992 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2993 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2995 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
2999 forget_connection(csp->sfd);
3000 close_socket(csp->sfd);
3003 close_socket(csp->sfd);
3004 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3007 csp->flags &= ~CSP_FLAG_ACTIVE;
3013 /*********************************************************************
3015 * Function : server_thread
3017 * Description : We only exist to call `serve' in a threaded environment.
3020 * 1 : data = Current client state (buffers, headers, etc...)
3022 * Returns : Always 0.
3024 *********************************************************************/
3025 static int32 server_thread(void *data)
3027 serve((struct client_state *) data);
3034 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3035 /*********************************************************************
3039 * Description : Print usage info & exit.
3041 * Parameters : Pointer to argv[0] for identifying ourselves
3045 *********************************************************************/
3046 static void usage(const char *myname)
3048 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3052 #endif /* defined(unix) */
3055 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3056 #endif /* defined(unix) */
3057 "[--version] [configfile]\n"
3058 "Aborting\n", myname);
3063 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3066 #ifdef MUTEX_LOCKS_AVAILABLE
3067 /*********************************************************************
3069 * Function : privoxy_mutex_lock
3071 * Description : Locks a mutex.
3074 * 1 : mutex = The mutex to lock.
3076 * Returns : Void. May exit in case of errors.
3078 *********************************************************************/
3079 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3081 #ifdef FEATURE_PTHREAD
3082 int err = pthread_mutex_lock(mutex);
3085 if (mutex != &log_mutex)
3087 log_error(LOG_LEVEL_FATAL,
3088 "Mutex locking failed: %s.\n", strerror(err));
3093 EnterCriticalSection(mutex);
3094 #endif /* def FEATURE_PTHREAD */
3098 /*********************************************************************
3100 * Function : privoxy_mutex_unlock
3102 * Description : Unlocks a mutex.
3105 * 1 : mutex = The mutex to unlock.
3107 * Returns : Void. May exit in case of errors.
3109 *********************************************************************/
3110 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3112 #ifdef FEATURE_PTHREAD
3113 int err = pthread_mutex_unlock(mutex);
3116 if (mutex != &log_mutex)
3118 log_error(LOG_LEVEL_FATAL,
3119 "Mutex unlocking failed: %s.\n", strerror(err));
3124 LeaveCriticalSection(mutex);
3125 #endif /* def FEATURE_PTHREAD */
3129 /*********************************************************************
3131 * Function : privoxy_mutex_init
3133 * Description : Prepares a mutex.
3136 * 1 : mutex = The mutex to initialize.
3138 * Returns : Void. May exit in case of errors.
3140 *********************************************************************/
3141 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3143 #ifdef FEATURE_PTHREAD
3144 int err = pthread_mutex_init(mutex, 0);
3147 printf("Fatal error. Mutex initialization failed: %s.\n",
3152 InitializeCriticalSection(mutex);
3153 #endif /* def FEATURE_PTHREAD */
3155 #endif /* def MUTEX_LOCKS_AVAILABLE */
3157 /*********************************************************************
3159 * Function : initialize_mutexes
3161 * Description : Prepares mutexes if mutex support is available.
3165 * Returns : Void, exits in case of errors.
3167 *********************************************************************/
3168 static void initialize_mutexes(void)
3170 #ifdef MUTEX_LOCKS_AVAILABLE
3172 * Prepare global mutex semaphores
3174 privoxy_mutex_init(&log_mutex);
3175 privoxy_mutex_init(&log_init_mutex);
3176 privoxy_mutex_init(&connection_reuse_mutex);
3179 * XXX: The assumptions below are a bit naive
3180 * and can cause locks that aren't necessary.
3182 * For example older FreeBSD versions (< 6.x?)
3183 * have no gethostbyname_r, but gethostbyname is
3186 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3187 privoxy_mutex_init(&resolver_mutex);
3188 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3190 * XXX: should we use a single mutex for
3191 * localtime() and gmtime() as well?
3193 #ifndef HAVE_GMTIME_R
3194 privoxy_mutex_init(&gmtime_mutex);
3195 #endif /* ndef HAVE_GMTIME_R */
3197 #ifndef HAVE_LOCALTIME_R
3198 privoxy_mutex_init(&localtime_mutex);
3199 #endif /* ndef HAVE_GMTIME_R */
3202 privoxy_mutex_init(&rand_mutex);
3203 #endif /* ndef HAVE_RANDOM */
3204 #endif /* def MUTEX_LOCKS_AVAILABLE */
3208 /*********************************************************************
3212 * Description : Load the config file and start the listen loop.
3213 * This function is a lot more *sane* with the `load_config'
3214 * and `listen_loop' functions; although it stills does
3215 * a *little* too much for my taste.
3218 * 1 : argc = Number of parameters (including $0).
3219 * 2 : argv = Array of (char *)'s to the parameters.
3221 * Returns : 1 if : can't open config file, unrecognized directive,
3222 * stats requested in multi-thread mode, can't open the
3223 * log file, can't open the jar file, listen port is invalid,
3224 * any load fails, and can't bind port.
3226 * Else main never returns, the process must be signaled
3227 * to terminate execution. Or, on Windows, use the
3228 * "File", "Exit" menu option.
3230 *********************************************************************/
3232 int real_main(int argc, const char *argv[])
3234 int main(int argc, const char *argv[])
3238 unsigned int random_seed;
3240 struct passwd *pw = NULL;
3241 struct group *grp = NULL;
3244 char *pre_chroot_nslookup_to_load_resolver = NULL;
3251 #if !defined(_WIN32)
3259 * Parse the command line arguments
3261 * XXX: simply printing usage information in case of
3262 * invalid arguments isn't particularly user friendly.
3264 while (++argc_pos < argc)
3267 /* Check to see if the service must be installed or uninstalled */
3268 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3270 const char *pName = argv[argc_pos] + 9;
3273 exit( (install_service(pName)) ? 0 : 1 );
3275 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3277 const char *pName = argv[argc_pos] + 11;
3280 exit((uninstall_service(pName)) ? 0 : 1);
3282 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3284 bRunAsService = TRUE;
3285 w32_set_service_cwd();
3286 atexit(w32_service_exit_notify);
3289 #endif /* defined(_WIN32) */
3292 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3294 if (strcmp(argv[argc_pos], "--help") == 0)
3299 else if(strcmp(argv[argc_pos], "--version") == 0)
3301 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3307 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3312 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3314 if (++argc_pos == argc) usage(argv[0]);
3315 pidfile = strdup(argv[argc_pos]);
3318 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3320 if (++argc_pos == argc) usage(argv[argc_pos]);
3322 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3325 if (NULL == (grp = getgrnam(p)))
3327 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3331 if (NULL == (pw = getpwnam(argv[argc_pos])))
3333 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3336 if (p != NULL) *--p = '\0';
3339 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3341 if (++argc_pos == argc) usage(argv[0]);
3342 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3345 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3349 #endif /* defined(unix) */
3351 else if (argc_pos + 1 != argc)
3354 * This is neither the last command line
3355 * option, nor was it recognized before,
3356 * therefore it must be invalid.
3362 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3364 configfile = argv[argc_pos];
3367 } /* -END- while (more arguments) */
3370 if ( *configfile != '/' )
3372 char cwd[BUFFER_SIZE];
3374 size_t abs_file_size;
3376 /* make config-filename absolute here */
3377 if (NULL == getcwd(cwd, sizeof(cwd)))
3379 perror("failed to get current working directory");
3384 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3385 basedir = strdup(cwd);
3387 if (NULL == basedir ||
3388 NULL == (abs_file = malloc(abs_file_size)))
3390 perror("malloc failed");
3393 strlcpy(abs_file, basedir, abs_file_size);
3394 strlcat(abs_file, "/", abs_file_size );
3395 strlcat(abs_file, configfile, abs_file_size);
3396 configfile = abs_file;
3398 #endif /* defined unix */
3402 clients->next = NULL;
3404 /* XXX: factor out initialising after the next stable release. */
3407 #elif defined(_WIN32)
3411 /* Prepare mutexes if supported and necessary. */
3412 initialize_mutexes();
3414 /* Enable logging until further notice. */
3415 init_log_module(Argv[0]);
3417 random_seed = (unsigned int)time(NULL);
3419 srandom(random_seed);
3422 #endif /* ifdef HAVE_RANDOM */
3425 * Unix signal handling
3427 * Catch the abort, interrupt and terminate signals for a graceful exit
3428 * Catch the hangup signal so the errlog can be reopened.
3429 * Ignore the broken pipe signals (FIXME: Why?)
3431 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3434 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3435 const int ignored_signals[] = { SIGPIPE, 0 };
3437 for (idx = 0; catched_signals[idx] != 0; idx++)
3439 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3440 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3442 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3443 #endif /* ifdef sun */
3445 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3449 for (idx = 0; ignored_signals[idx] != 0; idx++)
3451 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3453 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3458 #else /* ifdef _WIN32 */
3459 # ifdef _WIN_CONSOLE
3461 * We *are* in a windows console app.
3462 * Print a verbose messages about FAQ's and such
3464 printf("%s", win32_blurb);
3465 # endif /* def _WIN_CONSOLE */
3466 #endif /* def _WIN32 */
3469 /* Initialize the CGI subsystem */
3470 cgi_init_error_messages();
3473 * If runnig on unix and without the --nodaemon
3474 * option, become a daemon. I.e. fork, detach
3475 * from tty and get process group leadership
3488 if ( pid < 0 ) /* error */
3493 else if ( pid != 0 ) /* parent */
3498 * must check for errors
3499 * child died due to missing files aso
3502 wpid = waitpid( pid, &status, WNOHANG );
3511 /* Should be more portable, but not as well tested */
3516 #else /* ndef __FreeBSD__ */
3518 #endif /* ndef __FreeBSD__ */
3519 fd = open("/dev/tty", O_RDONLY);
3522 /* no error check here */
3523 ioctl( fd, TIOCNOTTY,0 );
3528 * stderr (fd 2) will be closed later on, when the
3529 * log file has been parsed.
3536 } /* -END- if (!no_daemon) */
3539 * As soon as we have written the PID file, we can switch
3540 * to the user and group ID indicated by the --user option
3546 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3548 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3554 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3556 /* Read the time zone file from /etc before doing chroot. */
3558 if (NULL != pre_chroot_nslookup_to_load_resolver
3559 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3561 /* Initialize resolver library. */
3562 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3564 if (chroot(pw->pw_dir) < 0)
3566 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3570 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3573 if (setuid(pw->pw_uid))
3575 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3579 char putenv_dummy[64];
3581 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3582 if (putenv(putenv_dummy) != 0)
3584 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3587 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3588 if (putenv(putenv_dummy) != 0)
3590 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3596 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3599 #endif /* defined unix */
3602 /* This will be FALSE unless the command line specified --service
3606 /* Yup, so now we must attempt to establish a connection
3607 * with the service dispatcher. This will only work if this
3608 * process was launched by the service control manager to
3609 * actually run as a service. If this isn't the case, i've
3610 * known it take around 30 seconds or so for the call to return.
3613 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3614 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3616 /* Service has run, and at this point is now being stopped, so just return */
3621 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3623 /* An error occurred. Usually it's because --service was wrongly specified
3624 * and we were unable to connect to the Service Control Dispatcher because
3625 * it wasn't expecting us and is therefore not listening.
3627 * For now, just continue below to call the listen_loop function.
3630 #endif /* def _WIN32 */
3640 /*********************************************************************
3642 * Function : bind_port_helper
3644 * Description : Bind the listen port. Handles logging, and aborts
3648 * 1 : config = Privoxy configuration. Specifies port
3651 * Returns : Port that was opened.
3653 *********************************************************************/
3654 static jb_socket bind_port_helper(struct configuration_spec * config)
3659 if (config->haddr == NULL)
3661 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3666 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3667 config->hport, config->haddr);
3670 result = bind_port(config->haddr, config->hport, &bfd);
3677 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3678 "There may be another Privoxy or some other "
3679 "proxy running on port %d",
3680 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3681 config->hport, config->hport);
3684 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3685 "The hostname is not resolvable",
3686 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3689 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3690 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3693 /* shouldn't get here */
3694 return JB_INVALID_SOCKET;
3697 config->need_bind = 0;
3704 /* Without this simple workaround we get this compiler warning from _beginthread
3705 * warning C4028: formal parameter 1 different from declaration
3707 void w32_service_listen_loop(void *p)
3711 #endif /* def _WIN32 */
3714 /*********************************************************************
3716 * Function : listen_loop
3718 * Description : bind the listen port and enter a "FOREVER" listening loop.
3724 *********************************************************************/
3725 static void listen_loop(void)
3727 struct client_state *csp = NULL;
3729 struct configuration_spec * config;
3731 config = load_config();
3733 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3735 * XXX: Should be relocated once it no
3736 * longer needs to emit log messages.
3738 initialize_reusable_connections();
3739 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3741 bfd = bind_port_helper(config);
3743 #ifdef FEATURE_GRACEFUL_TERMINATION
3744 while (!g_terminate)
3749 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3750 while (waitpid(-1, NULL, WNOHANG) > 0)
3752 /* zombie children */
3754 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3757 * Free data that was used by died threads
3763 * Re-open the errlog after HUP signal
3765 if (received_hup_signal)
3767 init_error_log(Argv[0], config->logfile);
3768 received_hup_signal = 0;
3772 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3774 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3778 csp->flags |= CSP_FLAG_ACTIVE;
3779 csp->sfd = JB_INVALID_SOCKET;
3781 csp->config = config = load_config();
3783 if ( config->need_bind )
3786 * Since we were listening to the "old port", we will not see
3787 * a "listen" param change until the next IJB request. So, at
3788 * least 1 more request must be made for us to find the new
3789 * setting. I am simply closing the old socket and binding the
3792 * Which-ever is correct, we will serve 1 more page via the
3793 * old settings. This should probably be a "show-proxy-args"
3794 * request. This should not be a so common of an operation
3795 * that this will hurt people's feelings.
3800 bfd = bind_port_helper(config);
3803 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3805 if (!accept_connection(csp, bfd))
3807 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3820 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3823 #ifdef FEATURE_TOGGLE
3824 if (global_toggle_state)
3825 #endif /* def FEATURE_TOGGLE */
3827 csp->flags |= CSP_FLAG_TOGGLED_ON;
3830 if (run_loader(csp))
3832 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3833 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3837 if (block_acl(NULL,csp))
3839 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3840 close_socket(csp->cfd);
3844 #endif /* def FEATURE_ACL */
3846 /* add it to the list of clients */
3847 csp->next = clients->next;
3848 clients->next = csp;
3850 if (config->multi_threaded)
3854 /* this is a switch () statment in the C preprocessor - ugh */
3855 #undef SELECTED_ONE_OPTION
3857 /* Use Pthreads in preference to native code */
3858 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3859 #define SELECTED_ONE_OPTION
3861 pthread_t the_thread;
3862 pthread_attr_t attrs;
3864 pthread_attr_init(&attrs);
3865 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3866 errno = pthread_create(&the_thread, &attrs,
3867 (void * (*)(void *))serve, csp);
3868 child_id = errno ? -1 : 0;
3869 pthread_attr_destroy(&attrs);
3873 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3874 #define SELECTED_ONE_OPTION
3875 child_id = _beginthread(
3876 (void (*)(void *))serve,
3881 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3882 #define SELECTED_ONE_OPTION
3883 child_id = _beginthread(
3884 (void(* _Optlink)(void*))serve,
3890 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3891 #define SELECTED_ONE_OPTION
3893 thread_id tid = spawn_thread
3894 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3896 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3898 child_id = (int) tid;
3907 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3908 #define SELECTED_ONE_OPTION
3909 csp->cfd = ReleaseSocket(csp->cfd, -1);
3912 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3913 NP_Output, Output(),
3914 NP_CloseOutput, FALSE,
3915 NP_Name, (ULONG)"privoxy child",
3919 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3920 NP_Output, Output(),
3921 NP_CloseOutput, FALSE,
3922 NP_Name, (ULONG)"privoxy child",
3923 NP_StackSize, 200*1024,
3929 ((struct Task *)child_id)->tc_UserData = csp;
3930 Signal((struct Task *)child_id, SIGF_SINGLE);
3935 #if !defined(SELECTED_ONE_OPTION)
3938 /* This block is only needed when using fork().
3939 * When using threads, the server thread was
3940 * created and run by the call to _beginthread().
3942 if (child_id == 0) /* child */
3945 #ifdef FEATURE_TOGGLE
3946 int inherited_toggle_state = global_toggle_state;
3947 #endif /* def FEATURE_TOGGLE */
3952 * If we've been toggled or we've blocked the request, tell Mom
3955 #ifdef FEATURE_TOGGLE
3956 if (inherited_toggle_state != global_toggle_state)
3958 rc |= RC_FLAG_TOGGLED;
3960 #endif /* def FEATURE_TOGGLE */
3962 #ifdef FEATURE_STATISTICS
3963 if (csp->flags & CSP_FLAG_REJECTED)
3965 rc |= RC_FLAG_BLOCKED;
3967 #endif /* ndef FEATURE_STATISTICS */
3971 else if (child_id > 0) /* parent */
3973 /* in a fork()'d environment, the parent's
3974 * copy of the client socket and the CSP
3978 #if !defined(_WIN32) && !defined(__CYGWIN__)
3980 wait( &child_status );
3983 * Evaluate child's return code: If the child has
3984 * - been toggled, toggle ourselves
3985 * - blocked its request, bump up the stats counter
3988 #ifdef FEATURE_TOGGLE
3989 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3991 global_toggle_state = !global_toggle_state;
3993 #endif /* def FEATURE_TOGGLE */
3995 #ifdef FEATURE_STATISTICS
3997 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4001 #endif /* def FEATURE_STATISTICS */
4003 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4004 close_socket(csp->cfd);
4005 csp->flags &= ~CSP_FLAG_ACTIVE;
4009 #undef SELECTED_ONE_OPTION
4010 /* end of cpp switch () */
4012 if (child_id < 0) /* failed */
4014 char buf[BUFFER_SIZE];
4016 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
4018 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
4020 write_socket(csp->cfd, buf, strlen(buf));
4021 close_socket(csp->cfd);
4022 csp->flags &= ~CSP_FLAG_ACTIVE;
4033 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4035 /* Clean up. Aim: free all memory (no leaks) */
4036 #ifdef FEATURE_GRACEFUL_TERMINATION
4038 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4040 unload_current_config_file();
4041 unload_current_actions_file();
4042 unload_current_re_filterfile();
4043 #ifdef FEATURE_TRUST
4044 unload_current_trust_file();
4047 if (config->multi_threaded)
4054 } while ((clients->next != NULL) && (--i > 0));
4058 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4069 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4070 /* Cleanup - remove taskbar icon etc. */
4075 #endif /* FEATURE_GRACEFUL_TERMINATION */