1 const char jcc_rcs[] = "$Id: jcc.c,v 1.183 2008/08/21 07:09:35 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 1.183 2008/08/21 07:09:35 fabiankeil
37 * Accept Shoutcast responses again. Problem reported
38 * and fix suggested by Stefan in #2062860.
40 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
41 * Fix possible NULL-pointer dereference reported
42 * by din_a4 in #2003937. Pointy hat to me.
44 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
45 * Streamline sed()'s prototype and declare
46 * the header parse and add structures static.
48 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
49 * - Mark csp as immutable for send_crunch_response().
50 * - Fix comment spelling.
52 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
53 * Factor update_server_headers() out of sed(), ditch the
54 * first_run hack and make server_patterns_light static.
56 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
57 * Don't provide get_header() with the whole client state
58 * structure when it only needs access to csp->iob.
60 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
61 * Make the "read the rest of the headers" loop a bit more readable.
63 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
64 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
65 * - Remove useless comment.
67 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
68 * Fix comment grammar.
70 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
71 * Remove the pointless buffer in client_protocol_is_unsupported().
73 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
74 * Least-effort fix for bug #1821930 (reported by Lee):
75 * If the response doesn't look like HTTP,
76 * tell the client and log the problem.
78 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
79 * Don't pass the whole csp structure to flush_socket()
80 * when it only needs a file descriptor and a buffer.
82 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
83 * Remove kill-popups action.
85 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
86 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
88 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
89 * Remove the treat-forbidden-connects-like-blocks action. We now
90 * use the "blocked" page for forbidden CONNECT requests by default.
92 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
93 * Also use shiny new connect_port_is_forbidden() in jcc.c.
95 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
96 * Rename url_actions() to get_url_actions() and let it
97 * use the standard parameter ordering.
99 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
100 * Let forward_url() use the standard parameter ordering
101 * and mark its second parameter immutable.
103 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
104 * Remove the "Listening ... for local connections only" log message.
105 * Whether or not remote connections are able to reach Privoxy is up
106 * to the operating system.
108 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
109 * Prevent the log messages for CONNECT requests to unacceptable
110 * ports from printing the limit-connect argument as [null] if
111 * limit-connect hasn't been explicitly enabled.
113 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
114 * Make sure all console-mode apps get a usage() instance
116 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
117 * Reword NO_SERVER_DATA_RESPONSE to make it harder
118 * to misunderstand what the message is all about.
120 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
121 * Unbreak trustfile which previously didn't work without
122 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
124 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
125 * Plug memory leak. Spotted by Valgrind, triggered by
126 * Privoxy-Regression-Test feeding proxyfuzz.py.
128 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
129 * In the HTTP snipplets, refer to the client as client.
131 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
132 * Emit a log message when activating the MS IIS5 hack.
134 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
135 * Log the "weak randomization factor" warning only
136 * once for mingw32 and provide some more details.
138 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
139 * Initialize log module after initializing mutexes, future
140 * deadlocks in that code should now work cross-platform.
142 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
143 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
144 * Reported by Bert van Leeuwen in BR#1818808.
146 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
147 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
149 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
150 * When in daemon mode, close stderr after the configuration file has been
151 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
153 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
154 * - Fix a crash when parsing invalid requests whose first header
155 * is rejected by get_header(). Regression (re?)introduced
156 * in r1.143 by yours truly.
157 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
158 * client_cookie_adder() to make sure send-vanilla-wafer can be
159 * controlled through tags (and thus regression-tested).
161 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
162 * - Move get_filter_function() from jcc.c to filters.c
163 * so the filter functions can be static.
164 * - Don't bother filtering body-less responses.
166 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
167 * Execute content filters through execute_content_filter().
169 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
170 * Initialize req to NULL to make sure it's defined if the
171 * first read_socket() call fails. Reported by icmp30.
173 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
174 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
175 * extensive comments moved to user manual.
177 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
178 * Don't crash if a broken header filter wiped out the request line.
180 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
181 * Fix byte_count calculation in case of flushes
182 * and don't parse the server headers a second time.
184 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
185 * - If there's a connection problem after we already forwarded
186 * parts of the original content, just hang up. Fixes BR#1776724.
187 * - Fix warnings about unused code on mingw32.
188 * - In case of flushes, calculate the byte count
189 * less incorrectly (I think).
191 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
192 * Add some more prototypes for static functions.
194 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
195 * Comment out request_contains_null_bytes() until it's used again.
197 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
198 * #1763173 from Stefan Huehner: s@const static@static const@
199 * and declare some more functions static.
201 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
202 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
203 * and explain why they were rejected in the first place.
204 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
205 * CONNECT requests. The request line was missing.
206 * - Add two more XXX reminders as we don't have enough already.
208 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
209 * As Hal noticed, checking dispatch_cgi() as the last cruncher
210 * looks like a bug if CGI requests are blocked unintentionally,
211 * so don't do it unless the user enabled the new config option
212 * "allow-cgi-request-crunching".
214 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
215 * - Allow to rewrite the request destination behind the client's back.
216 * - Turn the weird-looking unconditional for loop that
217 * reads the client request into a conditional while loop.
218 * Move the stuff that only runs once out of the loop.
219 * - Move parts of chat(), server_content_type() and the
220 * necessary stuff to fix BR#1750917 into get_filter_function().
222 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
223 * Temporary workaround for BR#1730105.
225 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
226 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
227 * deadlocks and crashes on OpenBSD and possibly other OS with neither
228 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
229 * Thanks to Ralf Horstmann for report and solution.
231 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
232 * Add forward-override{} to change the forwarding settings through
233 * action sections. This is mainly interesting to forward different
234 * clients differently (for example based on User-Agent or request
237 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
238 * - Let usage() mention the --chroot parameter.
239 * - Use read_socket() consistently and always leave
240 * the last buffer byte alone, even in cases where
241 * null termination (currently) doesn't matter.
243 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
244 * - Fix config file loading on Unix if no config file is specified.
245 * Since r1.97 Privoxy would always interpret the last argument as
246 * config file, even if it's a valid command line option.
247 * - Abort in case of unrecognized command line options. Closes #1719696.
248 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
249 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
251 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
252 * - Don't rerun crunchers that only depend on the request URL.
253 * - Don't count redirects and CGI requests as "blocked requests".
255 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
256 * Support crunching based on tags created by server-header taggers.
258 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
259 * Make HTTP snippets static (again). Add a Content-Type for those
260 * with content so the browser doesn't guess it based on the URL.
262 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
263 * Move crunching and request line rebuilding out of chat().
265 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
266 * Introduce tags as alternative way to specify which
267 * actions apply to a request. At the moment tags can be
268 * created based on client and server headers.
270 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
271 * Don't CLF-log CONNECT requests twice.
273 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
274 * Log the source address for ACL-related connection drops.
276 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
277 * New config option: enforce-blocks.
279 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
280 * - Move null byte check into separate function.
281 * - Don't confuse the client with error pages
282 * if a CONNECT request was already confirmed.
284 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
285 * Speed up NULL byte escaping and only log the complete
286 * NULL byte requests with header debugging enabled.
288 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
289 * Answer requests that contain NULL bytes with
290 * a custom response instead of waiting for more
291 * data until the client eventually hangs up.
293 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
294 * - Move delivery and logging of crunched responses
295 * from chat() into send_crunch_response().
296 * - Display the reason for generating http_responses.
297 * - Log the content length for LOG_LEVEL_CLF correctly
298 * (still incorrect for some fixed responses).
299 * - Reword an incorrect comment about
300 * treat-forbidden-connects-like-blocks violating
302 * - Add some log messages.
304 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
305 * Move mutex initialization into separate
306 * function and exit in case of errors.
308 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
309 * - Start to reduce chat()'s line count and move
310 * parts of it into separate functions.
311 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
313 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
314 * - Add Proxy-Agent header to HTTP snippets that are
315 * supposed to reach HTTP clients only.
316 * - Made a few CONNECT log messages more descriptive.
317 * - Catch completely empty server responses (as seen
318 * with Tor's fake ".noconnect" top level domain).
319 * - Use shiny new "forwarding-failed" template for socks errors.
321 * Revision 1.118 2007/01/07 07:43:43 joergs
322 * AmigaOS4 support added.
324 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
325 * Added config option accept-intercepted-requests
326 * and disabled it by default.
328 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
329 * Reverted parts of my last commit
330 * to keep error handling working.
332 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
333 * Fixed gcc43 conversion warnings.
335 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
336 * Fix -pedantic ISO C warning about converting
337 * from function pointer to object pointer.
339 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
340 * Silence compiler warning I introduced with my last commit.
342 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
343 * Mutex protect rand() if POSIX threading
344 * is used, warn the user if that's not possible
345 * and stop using it on _WIN32 where it could
348 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
349 * Don't prevent core dumps by catching SIGABRT.
350 * It's rude and makes debugging unreasonable painful.
352 * Revision 1.110 2006/12/13 14:52:53 etresoft
353 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
355 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
356 * Privoxy is now able to run as intercepting
357 * proxy in combination with any packet filter
358 * that does the port redirection. The destination
359 * is extracted from the "Host:" header which
360 * should be available for nearly all requests.
362 * Moved HTTP snipplets into jcc.c.
363 * Added error message for gopher proxy requests.
365 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
366 * Only unlink the pidfile if it's actually used.
368 * Change order of interception checks to make
369 * it possible to block or redirect requests for
372 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
373 * Make pthread mutex locking more generic. Instead of
374 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
375 * and use mutex locking unless there is an _r function
376 * available. Better safe than sorry.
378 * Fixes "./configure --disable-pthread" and should result
379 * in less threading-related problems on pthread-using platforms,
380 * but it still doesn't fix BR#1122404.
382 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
383 * Move pthread.h inclusion from jcc.c to jcc.h.
384 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
386 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
387 * Don't exit after receiving the second SIGHUP on Solaris.
389 * Fixes BR 1052235, but the same problem may exist on other
390 * systems. Once 3.0.6 is out we should use sigset()
391 * where available and see if it breaks anything.
393 * Revision 1.104 2006/09/23 13:26:38 roro
394 * Replace TABs by spaces in source code.
396 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
397 * Fix +redirect{}. Didn't work with -fast-redirects.
399 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
400 * Respond with 400 and a short text message
401 * if the client tries to use Privoxy as FTP proxy.
403 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
404 * Make number of retries in case of forwarded-connect problems
405 * a config file option (forwarded-connect-retries) and use 0 as
408 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
409 * Set random(3) seed.
411 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
412 * Follow the OpenBSD port's lead and protect the resolve
413 * functions on OpenBSD as well.
415 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
416 * --user fix. Only use the user as group if no group is specified.
417 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
419 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
420 * Windows service (re-)integration
424 * --install[:service_name]
425 * --uninstall[:service_name]
428 * They work as follows:
429 * --install will create a service for you and then terminate.
430 * By default the service name will be "privoxy" (without the quotes).
431 * However you can run multiple services if you wish, just by adding
432 * a colon and then a name (no spaces).
434 * --uninstall follows the exact same rules a --install.
436 * --service is used when the program is executed by the service
437 * control manager, and in normal circumstances would never be
438 * used as a command line argument.
440 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
441 * Windows service integration
443 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
444 * Incorporate Fabian Keil's patch work:
445 http://www.fabiankeil.de/sourcecode/privoxy/
447 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
448 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
449 * with what was really the latest development (the v_3_0_branch branch)
451 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
452 * Thanks to Jindrich Makovicka for a race condition fix for the log
453 * file. The race condition remains for non-pthread implementations.
454 * Reference patch #1175720.
456 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
457 * Add the ability to check jpeg images for invalid
458 * lengths of comment blocks. Defensive strategy
459 * against the exploit:
460 * Microsoft Security Bulletin MS04-028
461 * Buffer Overrun in JPEG Processing (GDI+) Could
462 * Allow Code Execution (833987)
463 * Enabled with +inspect-jpegs in actions files.
465 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
466 * - Fixed usage info for non-unix platforms
467 * - Fixed small cmdline parsing bug
469 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
470 * Diagnostics: Now preserve the returncode of pthread_create
471 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
473 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
474 * No longer ignore SIGCHLD. Fixes bug #769381
476 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
477 * Close jarfile on graceful exit, remove stray line
479 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
480 * Cosmetics: Killed a warning, a typo and an allocation left at exit
482 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
483 * No longer rely on non-POSIX.1 extensions of getcwd().
486 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
487 * Replaced setenv() by posix-compliant putenv()
488 * Thanks to Neil McCalden (nmcc AT users.sf.net).
490 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
491 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
493 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
494 * Clean-up and extension of improvements for forked mode:
495 * - Child's return code now consists of flags RC_FLAG_*
496 * - Reporting toggle to parent now properly #ifdef'ed
497 * - Children now report blocking to parent. This enables
498 * statistics in forked mode
500 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
501 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
502 * Children now report having been toggled through _exit(17), parents
503 * watch for that code and toggle themselves if found.
505 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
506 * Wrapping all *_r functions (the non-_r versions of them) with
507 * mutex semaphores for OSX. Hopefully this will take care of all
508 * of those pesky crash reports.
510 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
511 * Fixed two mostly harmless mem leaks
513 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
514 * Fix: Head of global clients list now initialized to NULL
516 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
517 * Added basic support for OPTIONS and TRACE HTTP methods:
518 * - New interceptor direct_response() added in chat().
519 * - sed() moved to earlier in the process, so that the
520 * Host: header is evaluated before actions and forwarding
523 * Revision 1.92 2002/05/08 16:00:46 oes
524 * Chat's buffer handling:
525 * - Fixed bug with unchecked out-of-mem conditions
526 * while reading client request & server headers
527 * - No longer predict if the buffer limit will be exceeded
528 * in the next read -- check add_to_iob's new
529 * return code. If buffer couldn't be extended
530 * (policy or out-of-mem) while
531 * - reading from client: abort
532 * - reading server headers: send error page
533 * - buffering server body for filter: flush,
534 * and if that fails: send error page
536 * Revision 1.91 2002/04/08 20:35:58 swa
539 * Revision 1.90 2002/04/02 14:57:28 oes
540 * Made sending wafers independent of FEATURE_COOKIE_JAR
542 * Revision 1.89 2002/03/31 17:18:59 jongfoster
543 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
545 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
546 * More compiler warning message maintenance
548 * Revision 1.87 2002/03/26 22:29:54 swa
549 * we have a new homepage!
551 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
552 * Workaround for closing the jarfile before load_config() comes around again
554 * Revision 1.85 2002/03/24 15:23:33 jongfoster
557 * Revision 1.84 2002/03/24 13:25:43 swa
558 * name change related issues
560 * Revision 1.83 2002/03/16 23:54:06 jongfoster
561 * Adding graceful termination feature, to help look for memory leaks.
562 * If you enable this (which, by design, has to be done by hand
563 * editing config.h) and then go to http://i.j.b/die, then the program
564 * will exit cleanly after the *next* request. It should free all the
565 * memory that was used.
567 * Revision 1.82 2002/03/13 00:27:05 jongfoster
570 * Revision 1.81 2002/03/12 01:42:50 oes
571 * Introduced modular filters
573 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
574 * OS/2 port maintenance:
575 * - Fixed EMX build - it had decayed a little
576 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
577 * substituted a memset for now.
579 * Revision 1.79 2002/03/09 20:03:52 jongfoster
580 * - Making various functions return int rather than size_t.
581 * (Undoing a recent change). Since size_t is unsigned on
582 * Windows, functions like read_socket that return -1 on
583 * error cannot return a size_t.
585 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
586 * crashes, and also frequently caused JB to jump to 100%
587 * CPU and stay there. (Because it thought it had just
588 * read ((unsigned)-1) == 4Gb of data...)
590 * - The signature of write_socket has changed, it now simply
591 * returns success=0/failure=nonzero.
593 * - Trying to get rid of a few warnings --with-debug on
594 * Windows, I've introduced a new type "jb_socket". This is
595 * used for the socket file descriptors. On Windows, this
596 * is SOCKET (a typedef for unsigned). Everywhere else, it's
597 * an int. The error value can't be -1 any more, so it's
598 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
599 * Windows it maps to the #define INVALID_SOCKET.)
601 * - The signature of bind_port has changed.
603 * Revision 1.78 2002/03/08 21:35:04 oes
604 * Added optional group supplement to --user option. Will now use default group of user if no group given
606 * Revision 1.77 2002/03/07 03:52:06 oes
607 * - Fixed compiler warnings etc
608 * - Improved handling of failed DNS lookups
610 * Revision 1.76 2002/03/06 22:54:35 jongfoster
611 * Automated function-comment nitpicking.
613 * Revision 1.75 2002/03/06 10:02:19 oes
614 * Fixed stupid bug when --user was not given
616 * Revision 1.74 2002/03/06 00:49:31 jongfoster
617 * Fixing warning on Windows
618 * Making #ifdefs that refer to the same variable consistently
619 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
621 * Revision 1.73 2002/03/05 23:57:30 hal9
622 * Stray character 's' on line 1618 was breaking build.
624 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
625 * - Re-enable OS/2 building after new parms were added
626 * - Fix false out of memory report when resolving CGI templates when no IP
627 * address is available of failed attempt (a la no such domain)
629 * Revision 1.71 2002/03/05 18:13:56 oes
630 * Added --user option
632 * Revision 1.70 2002/03/05 04:52:42 oes
633 * Deleted non-errlog debugging code
635 * Revision 1.69 2002/03/04 23:50:00 jongfoster
636 * Splitting off bind_port() call into bind_port_helper(), with
639 * Revision 1.68 2002/03/04 20:17:32 oes
642 * Revision 1.67 2002/03/04 18:18:57 oes
643 * - Removed _DEBUG mode
644 * - Cleand up cmdline parsing
645 * - Introduced --no-daemon, --pidfile options
646 * - Cleaned up signal handling:
647 * - Terminate cleanly on INT, TERM and ABRT
648 * - Schedule logfile for re-opening on HUP
649 * - Ignore CHLD and PIPE
650 * - Leave the rest with their default handlers
651 * - Uniform handler registration
652 * - Added usage() function
653 * - Played styleguide police
655 * Revision 1.66 2002/03/03 15:06:55 oes
656 * Re-enabled automatic config reloading
658 * Revision 1.65 2002/03/03 14:49:11 oes
659 * Fixed CLF logging: Now uses client's original HTTP request
661 * Revision 1.64 2002/03/03 09:18:03 joergs
662 * Made jumbjuster work on AmigaOS again.
664 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
665 * Clean up a little CRLF unpleasantness that suddenly appeared
667 * Revision 1.62 2002/02/20 23:17:23 jongfoster
668 * Detecting some out-of memory conditions and exiting with a log message.
670 * Revision 1.61 2002/01/17 21:01:52 jongfoster
671 * Moving all our URL and URL pattern parsing code to urlmatch.c.
673 * Revision 1.60 2001/12/30 14:07:32 steudten
674 * - Add signal handling (unix)
675 * - Add SIGHUP handler (unix)
676 * - Add creation of pidfile (unix)
677 * - Add action 'top' in rc file (RH)
678 * - Add entry 'SIGNALS' to manpage
679 * - Add exit message to logfile (unix)
681 * Revision 1.59 2001/12/13 14:07:18 oes
682 * Fixed Bug: 503 error page now sent OK
684 * Revision 1.58 2001/11/30 23:37:24 jongfoster
685 * Renaming the Win32 config file to config.txt - this is almost the
686 * same as the corresponding UNIX name "config"
688 * Revision 1.57 2001/11/16 00:47:43 jongfoster
689 * Changing the tty-disconnection code to use setsid().
691 * Revision 1.56 2001/11/13 20:20:54 jongfoster
692 * Tabs->spaces, fixing a bug with missing {} around an if()
694 * Revision 1.55 2001/11/13 20:14:53 jongfoster
695 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
697 * Revision 1.54 2001/11/07 00:03:14 steudten
698 * Give reliable return value if an error
699 * occurs not just 0 with new daemon mode.
701 * Revision 1.53 2001/11/05 21:41:43 steudten
702 * Add changes to be a real daemon just for unix os.
703 * (change cwd to /, detach from controlling tty, set
704 * process group and session leader to the own process.
706 * Add some fatal-error log message for failed malloc().
707 * Add '-d' if compiled with 'configure --with-debug' to
708 * enable debug output.
710 * Revision 1.52 2001/10/26 20:11:20 jongfoster
711 * Fixing type mismatch
713 * Revision 1.51 2001/10/26 17:38:28 oes
716 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
717 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
718 * threads to call select() simultaneously. So, it's time to do a real, live,
719 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
720 * (native). Both versions will work, but using __OS2__ offers multi-threading.
722 * Revision 1.49 2001/10/23 21:41:35 jongfoster
723 * Added call to initialize the (statically-allocated of course)
724 * "out of memory" CGI response.
726 * Revision 1.48 2001/10/10 19:56:46 jongfoster
727 * Moving some code that wasn't cookie-related out of an #ifdef
730 * Revision 1.47 2001/10/10 16:44:36 oes
731 * Added CONNECT destination port limitation check
733 * Revision 1.46 2001/10/08 15:17:41 oes
734 * Re-enabled SSL forwarding
736 * Revision 1.45 2001/10/07 15:42:11 oes
737 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
739 * Moved downgrading of the HTTP version from parse_http_request to
740 * chat(), since we can't decide if it is necessary before we have
741 * determined the actions for the URL. The HTTP command is now
742 * *always* re-built so the repairs need no longer be special-cased.
744 * filter_popups now gets a csp pointer so it can raise the new
745 * CSP_FLAG_MODIFIED flag.
749 * Added configurable size limit for the IOB. If the IOB grows so
750 * large that the next read would exceed the limit, the header
751 * is generated, and the header & unfiltered buffer are flushed
752 * to the client. Chat then continues in non-buffering,
753 * non-filtering body mode.
755 * Revision 1.44 2001/10/02 18:13:57 oes
758 * Revision 1.43 2001/10/02 15:32:13 oes
759 * Moved generation of hdr
761 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
762 * Cleaning up 2 compiler warnings on OS/2.
764 * Revision 1.41 2001/09/16 17:05:14 jongfoster
765 * Removing unused #include showarg.h
767 * Revision 1.40 2001/09/16 15:41:45 jongfoster
768 * Fixing signed/unsigned comparison warning.
770 * Revision 1.39 2001/09/16 13:21:27 jongfoster
771 * Changes to use new list functions.
773 * Revision 1.38 2001/09/16 13:01:46 jongfoster
774 * Removing redundant function call that zeroed zalloc()'d memory.
776 * Revision 1.37 2001/09/10 11:12:24 oes
777 * Deleted unused variable
779 * Revision 1.36 2001/09/10 10:56:15 oes
780 * Silenced compiler warnings
782 * Revision 1.35 2001/07/31 14:44:22 oes
783 * Deleted unused size parameter from filter_popups()
785 * Revision 1.34 2001/07/30 22:08:36 jongfoster
786 * Tidying up #defines:
787 * - All feature #defines are now of the form FEATURE_xxx
788 * - Permanently turned off WIN_GUI_EDIT
789 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
791 * Revision 1.33 2001/07/29 19:32:00 jongfoster
792 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
794 * Revision 1.32 2001/07/29 18:47:05 jongfoster
795 * Adding missing #include "loadcfg.h"
797 * Revision 1.31 2001/07/29 12:17:48 oes
798 * Applied pthread fix by Paul Lieverse
800 * Revision 1.30 2001/07/25 22:57:13 jongfoster
801 * __BEOS__ no longer overrides FEATURE_PTHREAD.
802 * This is because FEATURE_PTHREAD will soon be widely used, so I
803 * want to keep it simple.
805 * Revision 1.29 2001/07/24 12:47:06 oes
806 * Applied BeOS support update by Eugenia
808 * Revision 1.28 2001/07/23 13:26:12 oes
809 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
811 * Revision 1.27 2001/07/19 19:09:47 haroon
812 * - Added code to take care of the situation where while processing the first
813 * server response (which includes the server header), after finding the end
814 * of the headers we were not looking past the end of the headers for
815 * content modification. I enabled it for filter_popups.
816 * Someone else should look to see if other similar operations should be
817 * done to the discarded portion of the buffer.
819 * Note 2001/07/20: No, the other content modification mechanisms will process
820 * the whole iob later anyway. --oes
822 * Revision 1.26 2001/07/18 12:31:36 oes
825 * Revision 1.25 2001/07/15 19:43:49 jongfoster
826 * Supports POSIX threads.
827 * Also removed some unused #includes.
829 * Revision 1.24 2001/07/13 14:00:40 oes
830 * - Generic content modification scheme:
831 * Each feature has its own applicability flag that is set
832 * from csp->action->flags.
833 * Replaced the "filtering" int flag , by a function pointer
834 * "content_filter" to the function that will do the content
835 * modification. If it is != NULL, the document will be buffered
836 * and processed through *content_filter, which must set
837 * csp->content_length and return a modified copy of the body
838 * or return NULL (on failiure).
839 * - Changed csp->is_text to the more generic bitmap csp->content_type
840 * which can currently take the valued CT_TEXT or CT_GIF
842 * - Removed all #ifdef PCRS
844 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
845 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
846 * the `block_acl' call. This prevents a compilation error when the user
847 * does not wish to use the "ACL" feature.
849 * Revision 1.22 2001/06/29 21:45:41 oes
850 * Indentation, CRLF->LF, Tab-> Space
852 * Revision 1.21 2001/06/29 13:29:36 oes
853 * - Cleaned up, improved comments
854 * - Unified all possible interceptors (CGI,
855 * block, trust, fast_redirect) in one
856 * place, with one (CGI) answer generation
857 * mechansim. Much clearer now.
858 * - Removed the GIF image generation, which
859 * is now done in filters.c:block_url()
860 * - Made error conditions like domain lookup
861 * failiure or (various) problems while talking
862 * to the server use cgi.c:error_response()
863 * instead of generating HTML/HTTP in chat() (yuck!)
864 * - Removed logentry from cancelled commit
866 * Revision 1.20 2001/06/09 10:55:28 jongfoster
867 * Changing BUFSIZ ==> BUFFER_SIZE
869 * Revision 1.19 2001/06/07 23:12:52 jongfoster
870 * Replacing function pointer in struct gateway with a directly
871 * called function forwarded_connect().
872 * Replacing struct gateway with struct forward_spec
874 * Revision 1.18 2001/06/03 19:12:16 oes
875 * introduced new cgi handling
877 * Revision 1.17 2001/06/01 20:07:23 jongfoster
878 * Now uses action +image-blocker{} rather than config->tinygif
880 * Revision 1.16 2001/06/01 18:49:17 jongfoster
881 * Replaced "list_share" with "list" - the tiny memory gain was not
882 * worth the extra complexity.
884 * Revision 1.15 2001/05/31 21:24:47 jongfoster
885 * Changed "permission" to "action" throughout.
886 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
887 * Moved vanilla wafer check into chat(), since we must now
888 * decide whether or not to add it based on the URL.
890 * Revision 1.14 2001/05/29 20:14:01 joergs
891 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
894 * Revision 1.13 2001/05/29 09:50:24 jongfoster
895 * Unified blocklist/imagelist/permissionslist.
896 * File format is still under discussion, but the internal changes
899 * Also modified interceptor behaviour:
900 * - We now intercept all URLs beginning with one of the following
901 * prefixes (and *only* these prefixes):
903 * * http://ijbswa.sf.net/config/
904 * * http://ijbswa.sourceforge.net/config/
905 * - New interceptors "home page" - go to http://i.j.b/ to see it.
906 * - Internal changes so that intercepted and fast redirect pages
907 * are not replaced with an image.
908 * - Interceptors now have the option to send a binary page direct
909 * to the client. (i.e. ijb-send-banner uses this)
910 * - Implemented show-url-info interceptor. (Which is why I needed
911 * the above interceptors changes - a typical URL is
912 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
913 * The previous mechanism would not have intercepted that, and
914 * if it had been intercepted then it then it would have replaced
917 * Revision 1.12 2001/05/27 22:17:04 oes
919 * - re_process_buffer no longer writes the modified buffer
920 * to the client, which was very ugly. It now returns the
921 * buffer, which it is then written by chat.
923 * - content_length now adjusts the Content-Length: header
924 * for modified documents rather than crunch()ing it.
925 * (Length info in csp->content_length, which is 0 for
926 * unmodified documents)
928 * - For this to work, sed() is called twice when filtering.
930 * Revision 1.11 2001/05/26 17:27:53 jongfoster
931 * Added support for CLF and fixed LOG_LEVEL_LOG.
932 * Also did CRLF->LF fix of my previous patch.
934 * Revision 1.10 2001/05/26 15:26:15 jongfoster
935 * ACL feature now provides more security by immediately dropping
936 * connections from untrusted hosts.
938 * Revision 1.9 2001/05/26 00:28:36 jongfoster
939 * Automatic reloading of config file.
940 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
941 * Most of the global variables have been moved to a new
942 * struct configuration_spec, accessed through csp->config->globalname
943 * Most of the globals remaining are used by the Win32 GUI.
945 * Revision 1.8 2001/05/25 22:43:18 jongfoster
946 * Fixing minor memory leak and buffer overflow.
948 * Revision 1.7 2001/05/25 22:34:30 jongfoster
951 * Revision 1.6 2001/05/23 00:13:58 joergs
952 * AmigaOS support fixed.
954 * Revision 1.5 2001/05/22 18:46:04 oes
956 * - Enabled filtering banners by size rather than URL
957 * by adding patterns that replace all standard banner
958 * sizes with the "Junkbuster" gif to the re_filterfile
960 * - Enabled filtering WebBugs by providing a pattern
961 * which kills all 1x1 images
963 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
964 * which is selected by the (nonstandard and therefore
965 * capital) letter 'U' in the option string.
966 * It causes the quantifiers to be ungreedy by default.
967 * Appending a ? turns back to greedy (!).
969 * - Added a new interceptor ijb-send-banner, which
970 * sends back the "Junkbuster" gif. Without imagelist or
971 * MSIE detection support, or if tinygif = 1, or the
972 * URL isn't recognized as an imageurl, a lame HTML
973 * explanation is sent instead.
975 * - Added new feature, which permits blocking remote
976 * script redirects and firing back a local redirect
978 * The feature is conditionally compiled, i.e. it
979 * can be disabled with --disable-fast-redirects,
980 * plus it must be activated by a "fast-redirects"
981 * line in the config file, has its own log level
982 * and of course wants to be displayed by show-proxy-args
983 * Note: Boy, all the #ifdefs in 1001 locations and
984 * all the fumbling with configure.in and acconfig.h
985 * were *way* more work than the feature itself :-(
987 * - Because a generic redirect template was needed for
988 * this, tinygif = 3 now uses the same.
990 * - Moved GIFs, and other static HTTP response templates
995 * - Removed some >400 CRs again (Jon, you really worked
998 * Revision 1.4 2001/05/21 19:34:01 jongfoster
999 * Made failure to bind() a fatal error.
1001 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1002 * Version 2.9.4 checkin.
1003 * - Merged popupfile and cookiefile, and added control over PCRS
1004 * filtering, in new "permissionsfile".
1005 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1006 * file error you now get a message box (in the Win32 GUI) rather
1007 * than the program exiting with no explanation.
1008 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1010 * - Removed tabs from "config"
1011 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1012 * - Bumped up version number.
1014 * Revision 1.2 2001/05/17 22:34:44 oes
1015 * - Added hint on GIF char array generation to jcc.c
1016 * - Cleaned CRLF's from the sources and related files
1017 * - Repaired logging for REF and FRC
1019 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1020 * Initial import of version 2.9.3 source tree
1023 *********************************************************************/
1029 #include <sys/types.h>
1038 # ifndef FEATURE_PTHREAD
1042 # include <windows.h>
1043 # include <process.h>
1044 # endif /* ndef FEATURE_PTHREAD */
1047 # ifndef _WIN_CONSOLE
1048 # include "w32log.h"
1049 # endif /* ndef _WIN_CONSOLE */
1050 # include "w32svrapi.h"
1052 #else /* ifndef _WIN32 */
1054 # if !defined (__OS2__)
1055 # include <unistd.h>
1056 # include <sys/wait.h>
1057 # endif /* ndef __OS2__ */
1058 # include <sys/time.h>
1059 # include <sys/stat.h>
1060 # include <sys/ioctl.h>
1063 #include <sys/termios.h>
1071 # include <signal.h>
1074 # include <socket.h> /* BeOS has select() for sockets only. */
1075 # include <OS.h> /* declarations for threads and stuff. */
1078 # if defined(__EMX__) || defined(__OS2__)
1079 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1084 #define bzero(B,N) memset(B,0x00,n)
1088 # include <select.h>
1093 #include "project.h"
1096 #include "filters.h"
1097 #include "loaders.h"
1098 #include "parsers.h"
1099 #include "miscutil.h"
1101 #include "jbsockets.h"
1102 #include "gateway.h"
1103 #include "actions.h"
1105 #include "loadcfg.h"
1106 #include "urlmatch.h"
1108 const char jcc_h_rcs[] = JCC_H_VERSION;
1109 const char project_h_rcs[] = PROJECT_H_VERSION;
1112 struct client_state clients[1];
1113 struct file_list files[1];
1115 #ifdef FEATURE_STATISTICS
1116 int urls_read = 0; /* total nr of urls read inc rejected */
1117 int urls_rejected = 0; /* total nr of urls rejected */
1118 #endif /* def FEATURE_STATISTICS */
1120 #ifdef FEATURE_GRACEFUL_TERMINATION
1121 int g_terminate = 0;
1124 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1125 static void sig_handler(int the_signal);
1127 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1128 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1129 static jb_err get_server_headers(struct client_state *csp);
1130 static const char *crunch_reason(const struct http_response *rsp);
1131 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1133 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1135 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1136 static jb_err change_request_destination(struct client_state *csp);
1137 static void chat(struct client_state *csp);
1138 static void serve(struct client_state *csp);
1139 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1140 static void usage(const char *myname);
1142 static void initialize_mutexes(void);
1143 static jb_socket bind_port_helper(struct configuration_spec *config);
1144 static void listen_loop(void);
1147 void serve(struct client_state *csp);
1148 #else /* ifndef AMIGA */
1149 static void serve(struct client_state *csp);
1150 #endif /* def AMIGA */
1153 static int32 server_thread(void *data);
1154 #endif /* def __BEOS__ */
1157 #define sleep(N) Sleep(((N) * 1000))
1161 #define sleep(N) DosSleep(((N) * 100))
1164 #ifdef FEATURE_PTHREAD
1165 pthread_mutex_t log_mutex;
1166 pthread_mutex_t log_init_mutex;
1168 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1169 pthread_mutex_t resolver_mutex;
1170 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1172 #ifndef HAVE_GMTIME_R
1173 pthread_mutex_t gmtime_mutex;
1174 #endif /* ndef HAVE_GMTIME_R */
1176 #ifndef HAVE_LOCALTIME_R
1177 pthread_mutex_t localtime_mutex;
1178 #endif /* ndef HAVE_GMTIME_R */
1181 pthread_mutex_t rand_mutex;
1182 #endif /* ndef HAVE_RANDOM */
1184 #endif /* FEATURE_PTHREAD */
1187 const char *basedir = NULL;
1188 const char *pidfile = NULL;
1189 static int received_hup_signal = 0;
1190 #endif /* defined unix */
1192 /* HTTP snipplets. */
1193 static const char CSUCCEED[] =
1194 "HTTP/1.0 200 Connection established\r\n"
1195 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1197 static const char CHEADER[] =
1198 "HTTP/1.0 400 Invalid header received from client\r\n"
1199 "Proxy-Agent: Privoxy " VERSION "\r\n"
1200 "Content-Type: text/plain\r\n"
1201 "Connection: close\r\n\r\n"
1202 "Invalid header received from client.\r\n";
1204 static const char FTP_RESPONSE[] =
1205 "HTTP/1.0 400 Invalid request received from client\r\n"
1206 "Content-Type: text/plain\r\n"
1207 "Connection: close\r\n\r\n"
1208 "Invalid request. Privoxy doesn't support FTP.\r\n";
1210 static const char GOPHER_RESPONSE[] =
1211 "HTTP/1.0 400 Invalid request received from client\r\n"
1212 "Content-Type: text/plain\r\n"
1213 "Connection: close\r\n\r\n"
1214 "Invalid request. Privoxy doesn't support gopher.\r\n";
1216 /* XXX: should be a template */
1217 static const char MISSING_DESTINATION_RESPONSE[] =
1218 "HTTP/1.0 400 Bad request received from client\r\n"
1219 "Proxy-Agent: Privoxy " VERSION "\r\n"
1220 "Content-Type: text/plain\r\n"
1221 "Connection: close\r\n\r\n"
1222 "Bad request. Privoxy was unable to extract the destination.\r\n";
1224 /* XXX: should be a template */
1225 static const char NO_SERVER_DATA_RESPONSE[] =
1226 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1227 "Proxy-Agent: Privoxy " VERSION "\r\n"
1228 "Content-Type: text/plain\r\n"
1229 "Connection: close\r\n\r\n"
1230 "Empty server or forwarder response.\r\n"
1231 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1233 /* XXX: should be a template */
1234 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1235 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1236 "Proxy-Agent: Privoxy " VERSION "\r\n"
1237 "Content-Type: text/plain\r\n"
1238 "Connection: close\r\n\r\n"
1239 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1242 /* XXX: should be a template */
1243 static const char NULL_BYTE_RESPONSE[] =
1244 "HTTP/1.0 400 Bad request received from client\r\n"
1245 "Proxy-Agent: Privoxy " VERSION "\r\n"
1246 "Content-Type: text/plain\r\n"
1247 "Connection: close\r\n\r\n"
1248 "Bad request. Null byte(s) before end of request.\r\n";
1251 /* XXX: should be a template */
1252 static const char MESSED_UP_REQUEST_RESPONSE[] =
1253 "HTTP/1.0 400 Malformed request after rewriting\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. Messed up with header filters.\r\n";
1259 /* A function to crunch a response */
1260 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1262 /* Crunch function flags */
1263 #define CF_NO_FLAGS 0
1264 /* Cruncher applies to forced requests as well */
1265 #define CF_IGNORE_FORCE 1
1266 /* Crunched requests are counted for the block statistics */
1267 #define CF_COUNT_AS_REJECT 2
1269 /* A crunch function and its flags */
1272 const crunch_func_ptr cruncher;
1276 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1278 /* Complete list of cruncher functions */
1279 static const struct cruncher crunchers_all[] = {
1280 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1281 { block_url, CF_COUNT_AS_REJECT },
1282 #ifdef FEATURE_TRUST
1283 { trust_url, CF_COUNT_AS_REJECT },
1284 #endif /* def FEATURE_TRUST */
1285 { redirect_url, CF_NO_FLAGS },
1286 { dispatch_cgi, CF_IGNORE_FORCE},
1290 /* Light version, used after tags are applied */
1291 static const struct cruncher crunchers_light[] = {
1292 { block_url, CF_COUNT_AS_REJECT },
1293 { redirect_url, CF_NO_FLAGS },
1299 * XXX: Don't we really mean
1305 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1306 /*********************************************************************
1308 * Function : sig_handler
1310 * Description : Signal handler for different signals.
1311 * Exit gracefully on TERM and INT
1312 * or set a flag that will cause the errlog
1313 * to be reopened by the main thread on HUP.
1316 * 1 : the_signal = the signal cause this function to call
1320 *********************************************************************/
1321 static void sig_handler(int the_signal)
1327 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1339 received_hup_signal = 1;
1345 * We shouldn't be here, unless we catch signals
1346 * in main() that we can't handle here!
1348 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1356 /*********************************************************************
1358 * Function : client_protocol_is_unsupported
1360 * Description : Checks if the client used a known unsupported
1361 * protocol and deals with it by sending an error
1365 * 1 : csp = Current client state (buffers, headers, etc...)
1366 * 2 : req = the first request line send by the client
1368 * Returns : TRUE if an error response has been generated, or
1369 * FALSE if the request doesn't look invalid.
1371 *********************************************************************/
1372 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1375 * If it's a FTP or gopher request, we don't support it.
1377 * These checks are better than nothing, but they might
1378 * not work in all configurations and some clients might
1379 * have problems digesting the answer.
1381 * They should, however, never cause more problems than
1382 * Privoxy's old behaviour (returning the misleading HTML
1385 * "Could not resolve http://(ftp|gopher)://example.org").
1387 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1389 const char *response = NULL;
1390 const char *protocol = NULL;
1392 if (!strncmpic(req, "GET ftp://", 10))
1394 response = FTP_RESPONSE;
1399 response = GOPHER_RESPONSE;
1400 protocol = "GOPHER";
1402 log_error(LOG_LEVEL_ERROR,
1403 "%s tried to use Privoxy as %s proxy: %s",
1404 csp->ip_addr_str, protocol, req);
1405 log_error(LOG_LEVEL_CLF,
1406 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1408 write_socket(csp->cfd, response, strlen(response));
1417 /*********************************************************************
1419 * Function : get_request_destination_elsewhere
1421 * Description : If the client's request was redirected into
1422 * Privoxy without the client's knowledge,
1423 * the request line lacks the destination host.
1425 * This function tries to get it elsewhere,
1426 * provided accept-intercepted-requests is enabled.
1428 * "Elsewhere" currently only means "Host: header",
1429 * but in the future we may ask the redirecting
1430 * packet filter to look the destination up.
1432 * If the destination stays unknown, an error
1433 * response is send to the client and headers
1434 * are freed so that chat() can return directly.
1437 * 1 : csp = Current client state (buffers, headers, etc...)
1438 * 2 : headers = a header list
1440 * Returns : JB_ERR_OK if the destination is now known, or
1441 * JB_ERR_PARSE if it isn't.
1443 *********************************************************************/
1444 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1448 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1450 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1451 " Privoxy isn't configured to accept intercepted requests.",
1452 csp->ip_addr_str, csp->http->cmd);
1453 /* XXX: Use correct size */
1454 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1455 csp->ip_addr_str, csp->http->cmd);
1457 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1458 destroy_list(headers);
1460 return JB_ERR_PARSE;
1462 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1464 /* Split the domain we just got for pattern matching */
1465 init_domain_components(csp->http);
1471 /* We can't work without destination. Go spread the news.*/
1473 req = list_to_text(headers);
1475 /* XXX: Use correct size */
1476 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1477 csp->ip_addr_str, csp->http->cmd);
1478 log_error(LOG_LEVEL_ERROR,
1479 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1480 csp->ip_addr_str, csp->http->cmd, req);
1483 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1484 destroy_list(headers);
1486 return JB_ERR_PARSE;
1489 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1490 * to get the destination IP address, use it as host directly
1491 * or do a reverse DNS lookup first.
1496 /*********************************************************************
1498 * Function : get_server_headers
1500 * Description : Parses server headers in iob and fills them
1501 * into csp->headers so that they can later be
1505 * 1 : csp = Current client state (buffers, headers, etc...)
1507 * Returns : JB_ERR_OK if everything went fine, or
1508 * JB_ERR_PARSE if the headers were incomplete.
1510 *********************************************************************/
1511 static jb_err get_server_headers(struct client_state *csp)
1513 int continue_hack_in_da_house = 0;
1516 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1521 * continue hack in da house. Ignore the ending of
1522 * this head and continue enlisting header lines.
1523 * The reason is described below.
1525 enlist(csp->headers, "");
1526 continue_hack_in_da_house = 0;
1529 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1532 * It's a bodyless continue response, don't
1533 * stop header parsing after reaching its end.
1535 * As a result Privoxy will concatenate the
1536 * next response's head and parse and deliver
1537 * the headers as if they belonged to one request.
1539 * The client will separate them because of the
1540 * empty line between them.
1542 * XXX: What we're doing here is clearly against
1543 * the intended purpose of the continue header,
1544 * and under some conditions (HTTP/1.0 client request)
1545 * it's a standard violation.
1547 * Anyway, "sort of against the spec" is preferable
1548 * to "always getting confused by Continue responses"
1549 * (Privoxy's behaviour before this hack was added)
1551 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1552 continue_hack_in_da_house = 1;
1554 else if (*header == '\0')
1557 * If the header is empty, but the Continue hack
1558 * isn't active, we can assume that we reached the
1559 * end of the buffer before we hit the end of the
1562 * Inform the caller an let it decide how to handle it.
1564 return JB_ERR_PARSE;
1567 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1570 * XXX: Should we quit the request and return a
1571 * out of memory error page instead?
1573 log_error(LOG_LEVEL_ERROR,
1574 "Out of memory while enlisting server headers. %s lost.",
1584 /*********************************************************************
1586 * Function : crunch_reason
1588 * Description : Translates the crunch reason code into a string.
1591 * 1 : rsp = a http_response
1593 * Returns : A string with the crunch reason or an error description.
1595 *********************************************************************/
1596 static const char *crunch_reason(const struct http_response *rsp)
1598 char * reason = NULL;
1600 assert(rsp != NULL);
1603 return "Internal error while searching for crunch reason";
1606 switch (rsp->reason)
1608 case RSP_REASON_UNSUPPORTED:
1609 reason = "Unsupported HTTP feature";
1611 case RSP_REASON_BLOCKED:
1614 case RSP_REASON_UNTRUSTED:
1615 reason = "Untrusted";
1617 case RSP_REASON_REDIRECTED:
1618 reason = "Redirected";
1620 case RSP_REASON_CGI_CALL:
1621 reason = "CGI Call";
1623 case RSP_REASON_NO_SUCH_DOMAIN:
1624 reason = "DNS failure";
1626 case RSP_REASON_FORWARDING_FAILED:
1627 reason = "Forwarding failed";
1629 case RSP_REASON_CONNECT_FAILED:
1630 reason = "Connection failure";
1632 case RSP_REASON_OUT_OF_MEMORY:
1633 reason = "Out of memory (may mask other reasons)";
1636 reason = "No reason recorded";
1644 /*********************************************************************
1646 * Function : send_crunch_response
1648 * Description : Delivers already prepared response for
1649 * intercepted requests, logs the interception
1650 * and frees the response.
1653 * 1 : csp = Current client state (buffers, headers, etc...)
1654 * 1 : rsp = Fully prepared response. Will be freed on exit.
1656 * Returns : Nothing.
1658 *********************************************************************/
1659 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1661 const struct http_request *http = csp->http;
1662 char status_code[4];
1664 assert(rsp != NULL);
1665 assert(rsp->head != NULL);
1670 * Not supposed to happen. If it does
1671 * anyway, treat it as an unknown error.
1673 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1674 /* return code doesn't matter */
1679 /* If rsp is still NULL, we have serious internal problems. */
1680 log_error(LOG_LEVEL_FATAL,
1681 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1685 * Extract the status code from the actual head
1686 * that was send to the client. It is the only
1687 * way to get it right for all requests, including
1688 * the fixed ones for out-of-memory problems.
1690 * A head starts like this: 'HTTP/1.1 200...'
1694 status_code[0] = rsp->head[9];
1695 status_code[1] = rsp->head[10];
1696 status_code[2] = rsp->head[11];
1697 status_code[3] = '\0';
1699 /* Write the answer to the client */
1700 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1701 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1703 /* There is nothing we can do about it. */
1704 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1707 /* Log that the request was crunched and why. */
1708 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1709 http->hostport, http->path, crunch_reason(rsp));
1710 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1711 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1713 /* Clean up and return */
1714 if (cgi_error_memory() != rsp)
1716 free_http_response(rsp);
1723 /*********************************************************************
1725 * Function : request_contains_null_bytes
1727 * Description : Checks for NULL bytes in the request and sends
1728 * an error message to the client if any were found.
1730 * XXX: currently not used, see comment in chat().
1733 * 1 : csp = Current client state (buffers, headers, etc...)
1734 * 2 : buf = Data from the client's request to check.
1735 * 3 : len = The data length.
1737 * Returns : TRUE if the request contained one or more NULL bytes, or
1740 *********************************************************************/
1741 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1743 size_t c_len; /* Request lenght when treated as C string */
1745 c_len = strlen(buf);
1750 * Null byte(s) found. Log the request,
1751 * return an error response and hang up.
1753 size_t tmp_len = c_len;
1758 * Replace NULL byte(s) with '°' characters
1759 * so the request can be logged as string.
1760 * XXX: Is there a better replacement character?
1763 tmp_len += strlen(buf+tmp_len);
1764 } while (tmp_len < len);
1766 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1767 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1768 log_error(LOG_LEVEL_HEADER,
1769 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1771 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1773 /* XXX: Log correct size */
1774 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1784 /*********************************************************************
1786 * Function : crunch_response_triggered
1788 * Description : Checks if the request has to be crunched,
1789 * and delivers the crunch response if necessary.
1792 * 1 : csp = Current client state (buffers, headers, etc...)
1793 * 2 : crunchers = list of cruncher functions to run
1795 * Returns : TRUE if the request was answered with a crunch response
1798 *********************************************************************/
1799 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1801 struct http_response *rsp = NULL;
1802 const struct cruncher *c;
1805 * If CGI request crunching is disabled,
1806 * check the CGI dispatcher out of order to
1807 * prevent unintentional blocks or redirects.
1809 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1810 && (NULL != (rsp = dispatch_cgi(csp))))
1812 /* Deliver, log and free the interception response. */
1813 send_crunch_response(csp, rsp);
1817 for (c = crunchers; c->cruncher != NULL; c++)
1820 * Check the cruncher if either Privoxy is toggled
1821 * on and the request isn't forced, or if the cruncher
1822 * applies to forced requests as well.
1824 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1825 !(csp->flags & CSP_FLAG_FORCED)) ||
1826 (c->flags & CF_IGNORE_FORCE))
1828 rsp = c->cruncher(csp);
1831 /* Deliver, log and free the interception response. */
1832 send_crunch_response(csp, rsp);
1833 #ifdef FEATURE_STATISTICS
1834 if (c->flags & CF_COUNT_AS_REJECT)
1836 csp->flags |= CSP_FLAG_REJECTED;
1838 #endif /* def FEATURE_STATISTICS */
1849 /*********************************************************************
1851 * Function : build_request_line
1853 * Description : Builds the HTTP request line.
1855 * If a HTTP forwarder is used it expects the whole URL,
1856 * web servers only get the path.
1859 * 1 : csp = Current client state (buffers, headers, etc...)
1860 * 2 : fwd = The forwarding spec used for the request
1861 * XXX: Should use http->fwd instead.
1862 * 3 : request_line = The old request line which will be replaced.
1864 * Returns : Nothing. Terminates in case of memory problems.
1866 *********************************************************************/
1867 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1869 struct http_request *http = csp->http;
1871 assert(http->ssl == 0);
1874 * Downgrade http version from 1.1 to 1.0
1875 * if +downgrade action applies.
1877 if ( (csp->action->flags & ACTION_DOWNGRADE)
1878 && (!strcmpic(http->ver, "HTTP/1.1")))
1881 http->ver = strdup("HTTP/1.0");
1883 if (http->ver == NULL)
1885 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1890 * Rebuild the request line.
1892 freez(*request_line);
1893 *request_line = strdup(http->gpc);
1894 string_append(request_line, " ");
1896 if (fwd->forward_host)
1898 string_append(request_line, http->url);
1902 string_append(request_line, http->path);
1904 string_append(request_line, " ");
1905 string_append(request_line, http->ver);
1907 if (*request_line == NULL)
1909 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1911 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1915 /*********************************************************************
1917 * Function : change_request_destination
1919 * Description : Parse a (rewritten) request line and regenerate
1920 * the http request data.
1923 * 1 : csp = Current client state (buffers, headers, etc...)
1925 * Returns : Forwards the parse_http_request() return code.
1926 * Terminates in case of memory problems.
1928 *********************************************************************/
1929 static jb_err change_request_destination(struct client_state *csp)
1931 struct http_request *http = csp->http;
1934 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1935 free_http_request(http);
1936 err = parse_http_request(csp->headers->first->str, http, csp);
1937 if (JB_ERR_OK != err)
1939 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1940 jb_err_to_string(err));
1944 /* XXX: ocmd is a misleading name */
1945 http->ocmd = strdup(http->cmd);
1946 if (http->ocmd == NULL)
1948 log_error(LOG_LEVEL_FATAL,
1949 "Out of memory copying rewritten HTTP request line");
1957 /*********************************************************************
1961 * Description : Once a connection to the client has been accepted,
1962 * this function is called (via serve()) to handle the
1963 * main business of the communication. When this
1964 * function returns, the caller must close the client
1967 * FIXME: chat is nearly thousand lines long.
1971 * 1 : csp = Current client state (buffers, headers, etc...)
1973 * Returns : Nothing.
1975 *********************************************************************/
1976 static void chat(struct client_state *csp)
1978 char buf[BUFFER_SIZE];
1986 int ms_iis5_hack = 0;
1987 size_t byte_count = 0;
1988 int forwarded_connect_retries = 0;
1989 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1990 const struct forward_spec * fwd;
1991 struct http_request *http;
1992 int len; /* for buffer sizes (and negative error codes) */
1995 /* Function that does the content filtering for the current request */
1996 filter_function_ptr content_filter = NULL;
1998 /* Skeleton for HTTP response, if we should intercept the request */
1999 struct http_response *rsp;
2001 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2002 struct list header_list;
2003 struct list *headers = &header_list;
2007 memset(buf, 0, sizeof(buf));
2010 * Read the client's request. Note that since we're not using select() we
2011 * could get blocked here if a client connected, then didn't say anything!
2016 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2018 if (len <= 0) break; /* error! */
2021 * If there is no memory left for buffering the
2022 * request, there is nothing we can do but hang up
2024 if (add_to_iob(csp, buf, len))
2029 req = get_header(csp->iob);
2031 } while ((NULL != req) && ('\0' == *req));
2033 if ((NULL != req) && ('\0' != *req))
2035 /* Request received. Validate and parse it. */
2039 * XXX: Temporary disabled to prevent problems
2040 * with POST requests whose bodies are allowed to
2041 * contain NULL bytes. BR#1730105.
2043 * The main purpose of this check is to properly
2044 * log stuff like BitTorrent traffic and other junk
2045 * that hits public proxies. It's not required for
2046 * Privoxy to functions as those requests are discarded
2049 * It probably should be rewritten to only check
2050 * the head of the request. Another option would
2051 * be to let all POST requests pass, although that
2052 * may not be good enough.
2054 if (request_contains_null_bytes(csp, buf, len))
2056 /* NULL bytes found and dealt with, just hang up. */
2061 /* Does the request line look invalid? */
2062 if (client_protocol_is_unsupported(csp, req))
2065 * Yes. The request has already been
2066 * answered with a error response, the buffers
2067 * were freed and we're done with chatting.
2072 #ifdef FEATURE_FORCE_LOAD
2074 * If this request contains the FORCE_PREFIX and blocks
2075 * aren't enforced, get rid of it and set the force flag.
2077 if (strstr(req, FORCE_PREFIX))
2079 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2081 log_error(LOG_LEVEL_FORCE,
2082 "Ignored force prefix in request: \"%s\".", req);
2086 strclean(req, FORCE_PREFIX);
2087 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2088 csp->flags |= CSP_FLAG_FORCED;
2092 #endif /* def FEATURE_FORCE_LOAD */
2093 err = parse_http_request(req, http, csp);
2094 if (JB_ERR_OK != err)
2096 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2102 if (http->cmd == NULL)
2104 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2105 /* XXX: Use correct size */
2106 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2107 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2109 free_http_request(http);
2113 /* grab the rest of the client's headers */
2117 p = get_header(csp->iob);
2121 /* There are no additional headers to read. */
2128 * We didn't receive a complete header
2129 * line yet, get the rest of it.
2131 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2134 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2135 destroy_list(headers);
2139 if (add_to_iob(csp, buf, len))
2142 * If there is no memory left for buffering the
2143 * request, there is nothing we can do but hang up
2145 destroy_list(headers);
2152 * We were able to read a complete
2153 * header and can finaly enlist it.
2160 if (http->host == NULL)
2163 * If we still don't know the request destination,
2164 * the request is invalid or the client uses
2165 * Privoxy without its knowledge.
2167 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2170 * Our attempts to get the request destination
2171 * elsewhere failed or Privoxy is configured
2172 * to only accept proxy requests.
2174 * An error response has already been send
2175 * and we're done here.
2182 * Determine the actions for this URL
2184 #ifdef FEATURE_TOGGLE
2185 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2187 /* Most compatible set of actions (i.e. none) */
2188 init_current_action(csp->action);
2191 #endif /* ndef FEATURE_TOGGLE */
2193 get_url_actions(csp, http);
2197 * Save a copy of the original request for logging
2199 http->ocmd = strdup(http->cmd);
2201 if (http->ocmd == NULL)
2203 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2206 enlist(csp->headers, http->cmd);
2208 /* Append the previously read headers */
2209 list_append_list_unique(csp->headers, headers);
2210 destroy_list(headers);
2212 err = sed(csp, FILTER_CLIENT_HEADERS);
2213 if (JB_ERR_OK != err)
2215 assert(err == JB_ERR_PARSE);
2216 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2218 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2220 /* Check request line for rewrites. */
2221 if ((NULL == csp->headers->first->str)
2222 || (strcmp(http->cmd, csp->headers->first->str) &&
2223 (JB_ERR_OK != change_request_destination(csp))))
2226 * A header filter broke the request line - bail out.
2228 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2229 /* XXX: Use correct size */
2230 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2231 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2233 free_http_request(http);
2237 /* decide how to route the HTTP request */
2238 fwd = forward_url(csp, http);
2241 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2242 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2246 * build the http request to send to the server
2247 * we have to do one of the following:
2249 * create = use the original HTTP request to create a new
2250 * HTTP request that has either the path component
2251 * without the http://domainspec (w/path) or the
2252 * full orininal URL (w/url)
2253 * Note that the path and/or the HTTP version may
2254 * have been altered by now.
2256 * connect = Open a socket to the host:port of the server
2257 * and short-circuit server and client socket.
2259 * pass = Pass the request unchanged if forwarding a CONNECT
2260 * request to a parent proxy. Note that we'll be sending
2261 * the CFAIL message ourselves if connecting to the parent
2262 * fails, but we won't send a CSUCCEED message if it works,
2263 * since that would result in a double message (ours and the
2264 * parent's). After sending the request to the parent, we simply
2267 * here's the matrix:
2270 * +--------+--------+
2272 * 0 | create | connect|
2274 * Forwarding +--------+--------+
2276 * 1 | create | pass |
2278 * +--------+--------+
2282 if (http->ssl && connect_port_is_forbidden(csp))
2284 const char *acceptable_connect_ports =
2285 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2286 assert(NULL != acceptable_connect_ports);
2287 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2288 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2289 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2290 csp->action->flags |= ACTION_BLOCK;
2296 freez(csp->headers->first->str);
2297 build_request_line(csp, fwd, &csp->headers->first->str);
2300 hdr = list_to_text(csp->headers);
2303 /* FIXME Should handle error properly */
2304 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2308 * We have a request. Check if one of the crunchers wants it.
2310 if (crunch_response_triggered(csp, crunchers_all))
2313 * Yes. The client got the crunch response
2314 * and we are done here after cleaning up.
2317 list_remove_all(csp->headers);
2323 * The headers can't be removed earlier because
2324 * they were still needed for the referrer check
2325 * in case of CGI crunches.
2327 * XXX: Would it be worth to move the referrer check
2328 * into client_referrer() and set a flag if it's trusted?
2330 list_remove_all(csp->headers);
2332 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2334 if (fwd->forward_host)
2336 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2337 fwd->forward_host, fwd->forward_port, http->hostport);
2341 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2344 /* here we connect to the server, gateway, or the forwarder */
2346 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2347 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2349 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2350 forwarded_connect_retries, http->hostport);
2353 if (csp->sfd == JB_INVALID_SOCKET)
2355 if (fwd->type != SOCKS_NONE)
2358 rsp = error_response(csp, "forwarding-failed", errno);
2360 else if (errno == EINVAL)
2362 rsp = error_response(csp, "no-such-domain", errno);
2366 rsp = error_response(csp, "connect-failed", errno);
2367 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2372 /* Write the answer to the client */
2375 send_crunch_response(csp, rsp);
2382 if (fwd->forward_host || (http->ssl == 0))
2384 /* write the client's (modified) header to the server
2385 * (along with anything else that may be in the buffer)
2388 if (write_socket(csp->sfd, hdr, strlen(hdr))
2389 || (flush_socket(csp->sfd, csp->iob) < 0))
2391 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2394 rsp = error_response(csp, "connect-failed", errno);
2398 send_crunch_response(csp, rsp);
2408 * We're running an SSL tunnel and we're not forwarding,
2409 * so just send the "connect succeeded" message to the
2410 * client, flush the rest, and get out of the way.
2412 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2420 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2422 /* we're finished with the client's header */
2425 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2427 /* pass data between the client and server
2428 * until one or the other shuts down the connection.
2437 * FD_ZERO here seems to point to an errant macro which crashes.
2438 * So do this by hand for now...
2440 memset(&rfds,0x00,sizeof(fd_set));
2444 FD_SET(csp->cfd, &rfds);
2445 FD_SET(csp->sfd, &rfds);
2447 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2451 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2455 /* this is the body of the browser's request
2456 * just read it and write it.
2459 if (FD_ISSET(csp->cfd, &rfds))
2461 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2465 break; /* "game over, man" */
2468 if (write_socket(csp->sfd, buf, (size_t)len))
2470 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2477 * The server wants to talk. It could be the header or the body.
2478 * If `hdr' is null, then it's the header otherwise it's the body.
2479 * FIXME: Does `hdr' really mean `host'? No.
2483 if (FD_ISSET(csp->sfd, &rfds))
2486 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2490 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2492 if (http->ssl && (fwd->forward_host == NULL))
2495 * Just hang up. We already confirmed the client's CONNECT
2496 * request with status code 200 and unencrypted content is
2497 * no longer welcome.
2499 log_error(LOG_LEVEL_ERROR,
2500 "CONNECT already confirmed. Unable to tell the client about the problem.");
2503 else if (byte_count)
2506 * Just hang up. We already transmitted the original headers
2507 * and parts of the original content and therefore missed the
2508 * chance to send an error message (without risking data corruption).
2510 * XXX: we could retry with a fancy range request here.
2512 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2513 "Unable to tell the client about the problem.");
2517 rsp = error_response(csp, "connect-failed", errno);
2521 send_crunch_response(csp, rsp);
2527 /* Add a trailing zero. This lets filter_popups
2528 * use string operations.
2532 /* Normally, this would indicate that we've read
2533 * as much as the server has sent us and we can
2534 * close the client connection. However, Microsoft
2535 * in its wisdom has released IIS/5 with a bug that
2536 * prevents it from sending the trailing \r\n in
2537 * a 302 redirect header (and possibly other headers).
2538 * To work around this if we've haven't parsed
2539 * a full header we'll append a trailing \r\n
2540 * and see if this now generates a valid one.
2542 * This hack shouldn't have any impacts. If we've
2543 * already transmitted the header or if this is a
2544 * SSL connection, then we won't bother with this
2545 * hack. So we only work on partially received
2546 * headers. If we append a \r\n and this still
2547 * doesn't generate a valid header, then we won't
2548 * transmit anything to the client.
2553 if (server_body || http->ssl)
2556 * If we have been buffering up the document,
2557 * now is the time to apply content modification
2558 * and send the result to the client.
2562 p = execute_content_filter(csp, content_filter);
2564 * If the content filter fails, use the original
2565 * buffer and length.
2566 * (see p != NULL ? p : csp->iob->cur below)
2570 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2573 if (JB_ERR_OK != update_server_headers(csp))
2575 log_error(LOG_LEVEL_FATAL,
2576 "Failed to update server headers. after filtering.");
2579 hdr = list_to_text(csp->headers);
2582 /* FIXME Should handle error properly */
2583 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2586 if (write_socket(csp->cfd, hdr, strlen(hdr))
2587 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2589 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2599 break; /* "game over, man" */
2603 * This is NOT the body, so
2604 * Let's pretend the server just sent us a blank line.
2606 snprintf(buf, sizeof(buf), "\r\n");
2607 len = (int)strlen(buf);
2610 * Now, let the normal header parsing algorithm below do its
2611 * job. If it fails, we'll exit instead of continuing.
2618 * If this is an SSL connection or we're in the body
2619 * of the server document, just write it to the client,
2620 * unless we need to buffer the body for later content-filtering
2623 if (server_body || http->ssl)
2628 * If there is no memory left for buffering the content, or the buffer limit
2629 * has been reached, switch to non-filtering mode, i.e. make & write the
2630 * header, flush the iob and buf, and get out of the way.
2632 if (add_to_iob(csp, buf, len))
2637 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2639 hdr = list_to_text(csp->headers);
2643 * Memory is too tight to even generate the header.
2644 * Send our static "Out-of-memory" page.
2646 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2647 rsp = cgi_error_memory();
2648 send_crunch_response(csp, rsp);
2652 hdrlen = strlen(hdr);
2654 if (write_socket(csp->cfd, hdr, hdrlen)
2655 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2656 || (write_socket(csp->cfd, buf, (size_t)len)))
2658 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2665 * Reset the byte_count to the amount of bytes
2666 * we just flushed. len will be added a few lines below,
2667 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2669 byte_count = (size_t)flushed;
2671 content_filter = NULL;
2677 if (write_socket(csp->cfd, buf, (size_t)len))
2679 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2683 byte_count += (size_t)len;
2688 /* we're still looking for the end of the
2689 * server's header ... (does that make header
2690 * parsing an "out of body experience" ?
2694 * buffer up the data we just read. If that fails,
2695 * there's little we can do but send our static
2696 * out-of-memory page.
2698 if (add_to_iob(csp, buf, len))
2700 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2701 rsp = cgi_error_memory();
2702 send_crunch_response(csp, rsp);
2707 /* Convert iob into something sed() can digest */
2708 if (JB_ERR_PARSE == get_server_headers(csp))
2712 /* Well, we tried our MS IIS/5
2713 * hack and it didn't work.
2714 * The header is incomplete
2715 * and there isn't anything
2716 * we can do about it.
2718 log_error(LOG_LEVEL_INFO,
2719 "MS IIS5 hack didn't produce valid headers.");
2724 /* Since we have to wait for
2725 * more from the server before
2726 * we can parse the headers
2727 * we just continue here.
2733 /* Did we actually get anything? */
2734 if (NULL == csp->headers->first)
2736 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2737 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2738 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2739 free_http_request(http);
2743 assert(csp->headers->first->str);
2745 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2746 strncmpic(csp->headers->first->str, "ICY", 3))
2749 * It doesn't look like a HTTP (or Shoutcast) response:
2750 * tell the client and log the problem.
2752 if (strlen(csp->headers->first->str) > 30)
2754 csp->headers->first->str[30] = '\0';
2756 log_error(LOG_LEVEL_ERROR,
2757 "Invalid server or forwarder response. Starts with: %s",
2758 csp->headers->first->str);
2759 log_error(LOG_LEVEL_CLF,
2760 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2761 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2762 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2763 free_http_request(http);
2768 * We have now received the entire server header,
2769 * filter it and send the result to the client
2771 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2773 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2775 hdr = list_to_text(csp->headers);
2778 /* FIXME Should handle error properly */
2779 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2782 if (crunch_response_triggered(csp, crunchers_light))
2785 * One of the tags created by a server-header
2786 * tagger triggered a crunch. We already
2787 * delivered the crunch response to the client
2788 * and are done here after cleaning up.
2793 /* Buffer and pcrs filter this if appropriate. */
2795 if (!http->ssl) /* We talk plaintext */
2797 content_filter = get_filter_function(csp);
2800 * Only write if we're not buffering for content modification
2802 if (!content_filter)
2804 /* write the server's (modified) header to
2805 * the client (along with anything else that
2806 * may be in the buffer)
2809 if (write_socket(csp->cfd, hdr, strlen(hdr))
2810 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2812 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2814 /* the write failed, so don't bother
2815 * mentioning it to the client...
2816 * it probably can't hear us anyway.
2822 byte_count += (size_t)len;
2825 /* we're finished with the server's header */
2830 /* If this was a MS IIS/5 hack then it means
2831 * the server has already closed the
2832 * connection. Nothing more to read. Time
2837 log_error(LOG_LEVEL_INFO,
2838 "Closed server connection detected with MS IIS5 hack enabled.");
2845 return; /* huh? we should never get here */
2848 if (csp->content_length == 0)
2851 * If Privoxy didn't recalculate the
2852 * Content-Lenght, byte_count is still
2855 csp->content_length = byte_count;
2858 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2859 csp->ip_addr_str, http->ocmd, csp->content_length);
2863 /*********************************************************************
2867 * Description : This is little more than chat. We only "serve" to
2868 * to close any socket that chat may have opened.
2871 * 1 : csp = Current client state (buffers, headers, etc...)
2875 *********************************************************************/
2877 void serve(struct client_state *csp)
2878 #else /* ifndef AMIGA */
2879 static void serve(struct client_state *csp)
2880 #endif /* def AMIGA */
2883 close_socket(csp->cfd);
2885 if (csp->sfd != JB_INVALID_SOCKET)
2887 close_socket(csp->sfd);
2890 csp->flags &= ~CSP_FLAG_ACTIVE;
2896 /*********************************************************************
2898 * Function : server_thread
2900 * Description : We only exist to call `serve' in a threaded environment.
2903 * 1 : data = Current client state (buffers, headers, etc...)
2905 * Returns : Always 0.
2907 *********************************************************************/
2908 static int32 server_thread(void *data)
2910 serve((struct client_state *) data);
2917 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2918 /*********************************************************************
2922 * Description : Print usage info & exit.
2924 * Parameters : Pointer to argv[0] for identifying ourselves
2928 *********************************************************************/
2929 static void usage(const char *myname)
2931 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2935 #endif /* defined(unix) */
2938 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2939 #endif /* defined(unix) */
2940 "[--version] [configfile]\n"
2941 "Aborting\n", myname);
2946 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2949 /*********************************************************************
2951 * Function : initialize_mutexes
2953 * Description : Prepares mutexes if mutex support is available.
2957 * Returns : Void, exits in case of errors.
2959 *********************************************************************/
2960 static void initialize_mutexes(void)
2964 #ifdef FEATURE_PTHREAD
2966 * Prepare global mutex semaphores
2968 err = pthread_mutex_init(&log_mutex, 0);
2970 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2973 * XXX: The assumptions below are a bit naive
2974 * and can cause locks that aren't necessary.
2976 * For example older FreeBSD versions (< 6.x?)
2977 * have no gethostbyname_r, but gethostbyname is
2980 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2981 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2982 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2984 * XXX: should we use a single mutex for
2985 * localtime() and gmtime() as well?
2987 #ifndef HAVE_GMTIME_R
2988 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2989 #endif /* ndef HAVE_GMTIME_R */
2991 #ifndef HAVE_LOCALTIME_R
2992 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2993 #endif /* ndef HAVE_GMTIME_R */
2996 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2997 #endif /* ndef HAVE_RANDOM */
2998 #endif /* FEATURE_PTHREAD */
3001 * TODO: mutex support for mingw32 would be swell.
3006 printf("Fatal error. Mutex initialization failed: %s.\n",
3015 /*********************************************************************
3019 * Description : Load the config file and start the listen loop.
3020 * This function is a lot more *sane* with the `load_config'
3021 * and `listen_loop' functions; although it stills does
3022 * a *little* too much for my taste.
3025 * 1 : argc = Number of parameters (including $0).
3026 * 2 : argv = Array of (char *)'s to the parameters.
3028 * Returns : 1 if : can't open config file, unrecognized directive,
3029 * stats requested in multi-thread mode, can't open the
3030 * log file, can't open the jar file, listen port is invalid,
3031 * any load fails, and can't bind port.
3033 * Else main never returns, the process must be signaled
3034 * to terminate execution. Or, on Windows, use the
3035 * "File", "Exit" menu option.
3037 *********************************************************************/
3039 int real_main(int argc, const char *argv[])
3041 int main(int argc, const char *argv[])
3045 unsigned int random_seed;
3047 struct passwd *pw = NULL;
3048 struct group *grp = NULL;
3051 char *pre_chroot_nslookup_to_load_resolver = NULL;
3058 #if !defined(_WIN32)
3066 * Parse the command line arguments
3068 * XXX: simply printing usage information in case of
3069 * invalid arguments isn't particularly user friendly.
3071 while (++argc_pos < argc)
3074 /* Check to see if the service must be installed or uninstalled */
3075 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3077 const char *pName = argv[argc_pos] + 9;
3080 exit( (install_service(pName)) ? 0 : 1 );
3082 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3084 const char *pName = argv[argc_pos] + 11;
3087 exit((uninstall_service(pName)) ? 0 : 1);
3089 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3091 bRunAsService = TRUE;
3092 w32_set_service_cwd();
3093 atexit(w32_service_exit_notify);
3096 #endif /* defined(_WIN32) */
3099 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3101 if (strcmp(argv[argc_pos], "--help") == 0)
3106 else if(strcmp(argv[argc_pos], "--version") == 0)
3108 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3114 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3119 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3121 if (++argc_pos == argc) usage(argv[0]);
3122 pidfile = strdup(argv[argc_pos]);
3125 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3127 if (++argc_pos == argc) usage(argv[argc_pos]);
3129 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3132 if (NULL == (grp = getgrnam(p)))
3134 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3138 if (NULL == (pw = getpwnam(argv[argc_pos])))
3140 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3143 if (p != NULL) *--p = '\0';
3146 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3148 if (++argc_pos == argc) usage(argv[0]);
3149 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3152 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3156 #endif /* defined(unix) */
3158 else if (argc_pos + 1 != argc)
3161 * This is neither the last command line
3162 * option, nor was it recognized before,
3163 * therefore it must be invalid.
3169 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3171 configfile = argv[argc_pos];
3174 } /* -END- while (more arguments) */
3177 if ( *configfile != '/' )
3179 char cwd[BUFFER_SIZE];
3181 size_t abs_file_size;
3183 /* make config-filename absolute here */
3184 if (NULL == getcwd(cwd, sizeof(cwd)))
3186 perror("failed to get current working directory");
3191 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3192 basedir = strdup(cwd);
3194 if (NULL == basedir ||
3195 NULL == (abs_file = malloc(abs_file_size)))
3197 perror("malloc failed");
3200 strlcpy(abs_file, basedir, abs_file_size);
3201 strlcat(abs_file, "/", abs_file_size );
3202 strlcat(abs_file, configfile, abs_file_size);
3203 configfile = abs_file;
3205 #endif /* defined unix */
3209 clients->next = NULL;
3211 /* XXX: factor out initialising after the next stable release. */
3214 #elif defined(_WIN32)
3218 /* Prepare mutexes if supported and necessary. */
3219 initialize_mutexes();
3221 /* Enable logging until further notice. */
3222 init_log_module(Argv[0]);
3224 random_seed = (unsigned int)time(NULL);
3226 srandom(random_seed);
3227 #elif defined (_WIN32)
3229 * See pick_from_range() in miscutil.c for details.
3231 log_error(LOG_LEVEL_INFO,
3232 "No thread-safe PRNG implemented for your platform. "
3233 "Using weak \'randomization\' factor which will "
3234 "limit the already questionable usefulness of "
3235 "header-time-randomizing actions (disabled by default).");
3238 #endif /* ifdef HAVE_RANDOM */
3241 * Unix signal handling
3243 * Catch the abort, interrupt and terminate signals for a graceful exit
3244 * Catch the hangup signal so the errlog can be reopened.
3245 * Ignore the broken pipe signals (FIXME: Why?)
3247 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3250 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3251 const int ignored_signals[] = { SIGPIPE, 0 };
3253 for (idx = 0; catched_signals[idx] != 0; idx++)
3255 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3256 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3258 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3259 #endif /* ifdef sun */
3261 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3265 for (idx = 0; ignored_signals[idx] != 0; idx++)
3267 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3269 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3274 #else /* ifdef _WIN32 */
3275 # ifdef _WIN_CONSOLE
3277 * We *are* in a windows console app.
3278 * Print a verbose messages about FAQ's and such
3280 printf("%s", win32_blurb);
3281 # endif /* def _WIN_CONSOLE */
3282 #endif /* def _WIN32 */
3285 /* Initialize the CGI subsystem */
3286 cgi_init_error_messages();
3289 * If runnig on unix and without the --nodaemon
3290 * option, become a daemon. I.e. fork, detach
3291 * from tty and get process group leadership
3304 if ( pid < 0 ) /* error */
3309 else if ( pid != 0 ) /* parent */
3314 * must check for errors
3315 * child died due to missing files aso
3318 wpid = waitpid( pid, &status, WNOHANG );
3327 /* Should be more portable, but not as well tested */
3332 #else /* ndef __FreeBSD__ */
3334 #endif /* ndef __FreeBSD__ */
3335 fd = open("/dev/tty", O_RDONLY);
3338 /* no error check here */
3339 ioctl( fd, TIOCNOTTY,0 );
3344 * stderr (fd 2) will be closed later on, when the
3345 * log file has been parsed.
3352 } /* -END- if (!no_daemon) */
3355 * As soon as we have written the PID file, we can switch
3356 * to the user and group ID indicated by the --user option
3362 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3364 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3370 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3372 /* Read the time zone file from /etc before doing chroot. */
3374 if (NULL != pre_chroot_nslookup_to_load_resolver
3375 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3377 /* Initialize resolver library. */
3378 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3380 if (chroot(pw->pw_dir) < 0)
3382 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3386 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3389 if (setuid(pw->pw_uid))
3391 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3395 char putenv_dummy[64];
3397 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3398 if (putenv(putenv_dummy) != 0)
3400 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3403 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3404 if (putenv(putenv_dummy) != 0)
3406 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3412 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3415 #endif /* defined unix */
3418 /* This will be FALSE unless the command line specified --service
3422 /* Yup, so now we must attempt to establish a connection
3423 * with the service dispatcher. This will only work if this
3424 * process was launched by the service control manager to
3425 * actually run as a service. If this isn't the case, i've
3426 * known it take around 30 seconds or so for the call to return.
3429 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3430 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3432 /* Service has run, and at this point is now being stopped, so just return */
3437 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3439 /* An error occurred. Usually it's because --service was wrongly specified
3440 * and we were unable to connect to the Service Control Dispatcher because
3441 * it wasn't expecting us and is therefore not listening.
3443 * For now, just continue below to call the listen_loop function.
3446 #endif /* def _WIN32 */
3456 /*********************************************************************
3458 * Function : bind_port_helper
3460 * Description : Bind the listen port. Handles logging, and aborts
3464 * 1 : config = Privoxy configuration. Specifies port
3467 * Returns : Port that was opened.
3469 *********************************************************************/
3470 static jb_socket bind_port_helper(struct configuration_spec * config)
3475 if (config->haddr == NULL)
3477 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3482 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3483 config->hport, config->haddr);
3486 result = bind_port(config->haddr, config->hport, &bfd);
3493 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3494 "There may be another Privoxy or some other "
3495 "proxy running on port %d",
3496 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3497 config->hport, config->hport);
3500 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3501 "The hostname is not resolvable",
3502 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3505 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3506 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3509 /* shouldn't get here */
3510 return JB_INVALID_SOCKET;
3513 config->need_bind = 0;
3520 /* Without this simple workaround we get this compiler warning from _beginthread
3521 * warning C4028: formal parameter 1 different from declaration
3523 void w32_service_listen_loop(void *p)
3527 #endif /* def _WIN32 */
3530 /*********************************************************************
3532 * Function : listen_loop
3534 * Description : bind the listen port and enter a "FOREVER" listening loop.
3540 *********************************************************************/
3541 static void listen_loop(void)
3543 struct client_state *csp = NULL;
3545 struct configuration_spec * config;
3547 config = load_config();
3549 bfd = bind_port_helper(config);
3551 #ifdef FEATURE_GRACEFUL_TERMINATION
3552 while (!g_terminate)
3557 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3558 while (waitpid(-1, NULL, WNOHANG) > 0)
3560 /* zombie children */
3562 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3565 * Free data that was used by died threads
3571 * Re-open the errlog after HUP signal
3573 if (received_hup_signal)
3575 init_error_log(Argv[0], config->logfile);
3576 received_hup_signal = 0;
3581 #ifdef FEATURE_COOKIE_JAR
3583 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3584 * already open. I think unload_configfile() is not being run, which should do
3585 * this work. Until that can get resolved, we'll use this workaround.
3589 if (csp->config->jar)
3591 fclose(csp->config->jar);
3592 csp->config->jar = NULL;
3594 #endif /* FEATURE_COOKIE_JAR */
3595 #endif /* __OS2__ */
3597 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3599 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3603 csp->flags |= CSP_FLAG_ACTIVE;
3604 csp->sfd = JB_INVALID_SOCKET;
3606 csp->config = config = load_config();
3608 if ( config->need_bind )
3611 * Since we were listening to the "old port", we will not see
3612 * a "listen" param change until the next IJB request. So, at
3613 * least 1 more request must be made for us to find the new
3614 * setting. I am simply closing the old socket and binding the
3617 * Which-ever is correct, we will serve 1 more page via the
3618 * old settings. This should probably be a "show-proxy-args"
3619 * request. This should not be a so common of an operation
3620 * that this will hurt people's feelings.
3625 bfd = bind_port_helper(config);
3628 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3630 if (!accept_connection(csp, bfd))
3632 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3645 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3648 #ifdef FEATURE_TOGGLE
3649 if (global_toggle_state)
3650 #endif /* def FEATURE_TOGGLE */
3652 csp->flags |= CSP_FLAG_TOGGLED_ON;
3655 if (run_loader(csp))
3657 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3658 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3662 if (block_acl(NULL,csp))
3664 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3665 close_socket(csp->cfd);
3669 #endif /* def FEATURE_ACL */
3671 /* add it to the list of clients */
3672 csp->next = clients->next;
3673 clients->next = csp;
3675 if (config->multi_threaded)
3679 /* this is a switch () statment in the C preprocessor - ugh */
3680 #undef SELECTED_ONE_OPTION
3682 /* Use Pthreads in preference to native code */
3683 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3684 #define SELECTED_ONE_OPTION
3686 pthread_t the_thread;
3687 pthread_attr_t attrs;
3689 pthread_attr_init(&attrs);
3690 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3691 errno = pthread_create(&the_thread, &attrs,
3692 (void * (*)(void *))serve, csp);
3693 child_id = errno ? -1 : 0;
3694 pthread_attr_destroy(&attrs);
3698 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3699 #define SELECTED_ONE_OPTION
3700 child_id = _beginthread(
3701 (void (*)(void *))serve,
3706 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3707 #define SELECTED_ONE_OPTION
3708 child_id = _beginthread(
3709 (void(* _Optlink)(void*))serve,
3715 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3716 #define SELECTED_ONE_OPTION
3718 thread_id tid = spawn_thread
3719 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3721 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3723 child_id = (int) tid;
3732 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3733 #define SELECTED_ONE_OPTION
3734 csp->cfd = ReleaseSocket(csp->cfd, -1);
3737 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3738 NP_Output, Output(),
3739 NP_CloseOutput, FALSE,
3740 NP_Name, (ULONG)"privoxy child",
3744 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3745 NP_Output, Output(),
3746 NP_CloseOutput, FALSE,
3747 NP_Name, (ULONG)"privoxy child",
3748 NP_StackSize, 200*1024,
3754 ((struct Task *)child_id)->tc_UserData = csp;
3755 Signal((struct Task *)child_id, SIGF_SINGLE);
3760 #if !defined(SELECTED_ONE_OPTION)
3763 /* This block is only needed when using fork().
3764 * When using threads, the server thread was
3765 * created and run by the call to _beginthread().
3767 if (child_id == 0) /* child */
3770 #ifdef FEATURE_TOGGLE
3771 int inherited_toggle_state = global_toggle_state;
3772 #endif /* def FEATURE_TOGGLE */
3777 * If we've been toggled or we've blocked the request, tell Mom
3780 #ifdef FEATURE_TOGGLE
3781 if (inherited_toggle_state != global_toggle_state)
3783 rc |= RC_FLAG_TOGGLED;
3785 #endif /* def FEATURE_TOGGLE */
3787 #ifdef FEATURE_STATISTICS
3788 if (csp->flags & CSP_FLAG_REJECTED)
3790 rc |= RC_FLAG_BLOCKED;
3792 #endif /* ndef FEATURE_STATISTICS */
3796 else if (child_id > 0) /* parent */
3798 /* in a fork()'d environment, the parent's
3799 * copy of the client socket and the CSP
3803 #if !defined(_WIN32) && !defined(__CYGWIN__)
3805 wait( &child_status );
3808 * Evaluate child's return code: If the child has
3809 * - been toggled, toggle ourselves
3810 * - blocked its request, bump up the stats counter
3813 #ifdef FEATURE_TOGGLE
3814 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3816 global_toggle_state = !global_toggle_state;
3818 #endif /* def FEATURE_TOGGLE */
3820 #ifdef FEATURE_STATISTICS
3822 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3826 #endif /* def FEATURE_STATISTICS */
3828 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3829 close_socket(csp->cfd);
3830 csp->flags &= ~CSP_FLAG_ACTIVE;
3834 #undef SELECTED_ONE_OPTION
3835 /* end of cpp switch () */
3837 if (child_id < 0) /* failed */
3839 char buf[BUFFER_SIZE];
3841 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3843 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3845 write_socket(csp->cfd, buf, strlen(buf));
3846 close_socket(csp->cfd);
3847 csp->flags &= ~CSP_FLAG_ACTIVE;
3858 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3860 /* Clean up. Aim: free all memory (no leaks) */
3861 #ifdef FEATURE_GRACEFUL_TERMINATION
3863 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3865 unload_current_config_file();
3866 unload_current_actions_file();
3867 unload_current_re_filterfile();
3868 #ifdef FEATURE_TRUST
3869 unload_current_trust_file();
3872 if (config->multi_threaded)
3879 } while ((clients->next != NULL) && (--i > 0));
3883 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3894 #ifdef FEATURE_COOKIE_JAR
3895 if (NULL != config->jar)
3897 fclose(config->jar);
3901 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3902 /* Cleanup - remove taskbar icon etc. */
3907 #endif /* FEATURE_GRACEFUL_TERMINATION */