1 const char jcc_rcs[] = "$Id: jcc.c,v 1.185 2008/08/30 12:03:07 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 1.185 2008/08/30 12:03:07 fabiankeil
37 * Remove FEATURE_COOKIE_JAR.
39 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
40 * - Silence LLVM/Clang complaint.
41 * - Make received_hup_signal static.
42 * - Hide definitions for basedir, pidfile and received_hup_signal
43 * from __EMX__ as they only seem to be used in case of #ifdef unix.
45 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
46 * Accept Shoutcast responses again. Problem reported
47 * and fix suggested by Stefan in #2062860.
49 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
50 * Fix possible NULL-pointer dereference reported
51 * by din_a4 in #2003937. Pointy hat to me.
53 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
54 * Streamline sed()'s prototype and declare
55 * the header parse and add structures static.
57 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
58 * - Mark csp as immutable for send_crunch_response().
59 * - Fix comment spelling.
61 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
62 * Factor update_server_headers() out of sed(), ditch the
63 * first_run hack and make server_patterns_light static.
65 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
66 * Don't provide get_header() with the whole client state
67 * structure when it only needs access to csp->iob.
69 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
70 * Make the "read the rest of the headers" loop a bit more readable.
72 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
73 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
74 * - Remove useless comment.
76 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
77 * Fix comment grammar.
79 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
80 * Remove the pointless buffer in client_protocol_is_unsupported().
82 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
83 * Least-effort fix for bug #1821930 (reported by Lee):
84 * If the response doesn't look like HTTP,
85 * tell the client and log the problem.
87 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
88 * Don't pass the whole csp structure to flush_socket()
89 * when it only needs a file descriptor and a buffer.
91 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
92 * Remove kill-popups action.
94 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
95 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
97 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
98 * Remove the treat-forbidden-connects-like-blocks action. We now
99 * use the "blocked" page for forbidden CONNECT requests by default.
101 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
102 * Also use shiny new connect_port_is_forbidden() in jcc.c.
104 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
105 * Rename url_actions() to get_url_actions() and let it
106 * use the standard parameter ordering.
108 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
109 * Let forward_url() use the standard parameter ordering
110 * and mark its second parameter immutable.
112 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
113 * Remove the "Listening ... for local connections only" log message.
114 * Whether or not remote connections are able to reach Privoxy is up
115 * to the operating system.
117 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
118 * Prevent the log messages for CONNECT requests to unacceptable
119 * ports from printing the limit-connect argument as [null] if
120 * limit-connect hasn't been explicitly enabled.
122 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
123 * Make sure all console-mode apps get a usage() instance
125 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
126 * Reword NO_SERVER_DATA_RESPONSE to make it harder
127 * to misunderstand what the message is all about.
129 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
130 * Unbreak trustfile which previously didn't work without
131 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
133 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
134 * Plug memory leak. Spotted by Valgrind, triggered by
135 * Privoxy-Regression-Test feeding proxyfuzz.py.
137 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
138 * In the HTTP snipplets, refer to the client as client.
140 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
141 * Emit a log message when activating the MS IIS5 hack.
143 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
144 * Log the "weak randomization factor" warning only
145 * once for mingw32 and provide some more details.
147 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
148 * Initialize log module after initializing mutexes, future
149 * deadlocks in that code should now work cross-platform.
151 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
152 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
153 * Reported by Bert van Leeuwen in BR#1818808.
155 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
156 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
158 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
159 * When in daemon mode, close stderr after the configuration file has been
160 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
162 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
163 * - Fix a crash when parsing invalid requests whose first header
164 * is rejected by get_header(). Regression (re?)introduced
165 * in r1.143 by yours truly.
166 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
167 * client_cookie_adder() to make sure send-vanilla-wafer can be
168 * controlled through tags (and thus regression-tested).
170 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
171 * - Move get_filter_function() from jcc.c to filters.c
172 * so the filter functions can be static.
173 * - Don't bother filtering body-less responses.
175 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
176 * Execute content filters through execute_content_filter().
178 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
179 * Initialize req to NULL to make sure it's defined if the
180 * first read_socket() call fails. Reported by icmp30.
182 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
183 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
184 * extensive comments moved to user manual.
186 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
187 * Don't crash if a broken header filter wiped out the request line.
189 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
190 * Fix byte_count calculation in case of flushes
191 * and don't parse the server headers a second time.
193 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
194 * - If there's a connection problem after we already forwarded
195 * parts of the original content, just hang up. Fixes BR#1776724.
196 * - Fix warnings about unused code on mingw32.
197 * - In case of flushes, calculate the byte count
198 * less incorrectly (I think).
200 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
201 * Add some more prototypes for static functions.
203 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
204 * Comment out request_contains_null_bytes() until it's used again.
206 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
207 * #1763173 from Stefan Huehner: s@const static@static const@
208 * and declare some more functions static.
210 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
211 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
212 * and explain why they were rejected in the first place.
213 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
214 * CONNECT requests. The request line was missing.
215 * - Add two more XXX reminders as we don't have enough already.
217 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
218 * As Hal noticed, checking dispatch_cgi() as the last cruncher
219 * looks like a bug if CGI requests are blocked unintentionally,
220 * so don't do it unless the user enabled the new config option
221 * "allow-cgi-request-crunching".
223 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
224 * - Allow to rewrite the request destination behind the client's back.
225 * - Turn the weird-looking unconditional for loop that
226 * reads the client request into a conditional while loop.
227 * Move the stuff that only runs once out of the loop.
228 * - Move parts of chat(), server_content_type() and the
229 * necessary stuff to fix BR#1750917 into get_filter_function().
231 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
232 * Temporary workaround for BR#1730105.
234 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
235 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
236 * deadlocks and crashes on OpenBSD and possibly other OS with neither
237 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
238 * Thanks to Ralf Horstmann for report and solution.
240 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
241 * Add forward-override{} to change the forwarding settings through
242 * action sections. This is mainly interesting to forward different
243 * clients differently (for example based on User-Agent or request
246 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
247 * - Let usage() mention the --chroot parameter.
248 * - Use read_socket() consistently and always leave
249 * the last buffer byte alone, even in cases where
250 * null termination (currently) doesn't matter.
252 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
253 * - Fix config file loading on Unix if no config file is specified.
254 * Since r1.97 Privoxy would always interpret the last argument as
255 * config file, even if it's a valid command line option.
256 * - Abort in case of unrecognized command line options. Closes #1719696.
257 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
258 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
260 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
261 * - Don't rerun crunchers that only depend on the request URL.
262 * - Don't count redirects and CGI requests as "blocked requests".
264 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
265 * Support crunching based on tags created by server-header taggers.
267 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
268 * Make HTTP snippets static (again). Add a Content-Type for those
269 * with content so the browser doesn't guess it based on the URL.
271 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
272 * Move crunching and request line rebuilding out of chat().
274 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
275 * Introduce tags as alternative way to specify which
276 * actions apply to a request. At the moment tags can be
277 * created based on client and server headers.
279 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
280 * Don't CLF-log CONNECT requests twice.
282 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
283 * Log the source address for ACL-related connection drops.
285 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
286 * New config option: enforce-blocks.
288 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
289 * - Move null byte check into separate function.
290 * - Don't confuse the client with error pages
291 * if a CONNECT request was already confirmed.
293 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
294 * Speed up NULL byte escaping and only log the complete
295 * NULL byte requests with header debugging enabled.
297 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
298 * Answer requests that contain NULL bytes with
299 * a custom response instead of waiting for more
300 * data until the client eventually hangs up.
302 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
303 * - Move delivery and logging of crunched responses
304 * from chat() into send_crunch_response().
305 * - Display the reason for generating http_responses.
306 * - Log the content length for LOG_LEVEL_CLF correctly
307 * (still incorrect for some fixed responses).
308 * - Reword an incorrect comment about
309 * treat-forbidden-connects-like-blocks violating
311 * - Add some log messages.
313 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
314 * Move mutex initialization into separate
315 * function and exit in case of errors.
317 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
318 * - Start to reduce chat()'s line count and move
319 * parts of it into separate functions.
320 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
322 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
323 * - Add Proxy-Agent header to HTTP snippets that are
324 * supposed to reach HTTP clients only.
325 * - Made a few CONNECT log messages more descriptive.
326 * - Catch completely empty server responses (as seen
327 * with Tor's fake ".noconnect" top level domain).
328 * - Use shiny new "forwarding-failed" template for socks errors.
330 * Revision 1.118 2007/01/07 07:43:43 joergs
331 * AmigaOS4 support added.
333 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
334 * Added config option accept-intercepted-requests
335 * and disabled it by default.
337 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
338 * Reverted parts of my last commit
339 * to keep error handling working.
341 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
342 * Fixed gcc43 conversion warnings.
344 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
345 * Fix -pedantic ISO C warning about converting
346 * from function pointer to object pointer.
348 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
349 * Silence compiler warning I introduced with my last commit.
351 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
352 * Mutex protect rand() if POSIX threading
353 * is used, warn the user if that's not possible
354 * and stop using it on _WIN32 where it could
357 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
358 * Don't prevent core dumps by catching SIGABRT.
359 * It's rude and makes debugging unreasonable painful.
361 * Revision 1.110 2006/12/13 14:52:53 etresoft
362 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
364 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
365 * Privoxy is now able to run as intercepting
366 * proxy in combination with any packet filter
367 * that does the port redirection. The destination
368 * is extracted from the "Host:" header which
369 * should be available for nearly all requests.
371 * Moved HTTP snipplets into jcc.c.
372 * Added error message for gopher proxy requests.
374 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
375 * Only unlink the pidfile if it's actually used.
377 * Change order of interception checks to make
378 * it possible to block or redirect requests for
381 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
382 * Make pthread mutex locking more generic. Instead of
383 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
384 * and use mutex locking unless there is an _r function
385 * available. Better safe than sorry.
387 * Fixes "./configure --disable-pthread" and should result
388 * in less threading-related problems on pthread-using platforms,
389 * but it still doesn't fix BR#1122404.
391 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
392 * Move pthread.h inclusion from jcc.c to jcc.h.
393 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
395 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
396 * Don't exit after receiving the second SIGHUP on Solaris.
398 * Fixes BR 1052235, but the same problem may exist on other
399 * systems. Once 3.0.6 is out we should use sigset()
400 * where available and see if it breaks anything.
402 * Revision 1.104 2006/09/23 13:26:38 roro
403 * Replace TABs by spaces in source code.
405 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
406 * Fix +redirect{}. Didn't work with -fast-redirects.
408 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
409 * Respond with 400 and a short text message
410 * if the client tries to use Privoxy as FTP proxy.
412 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
413 * Make number of retries in case of forwarded-connect problems
414 * a config file option (forwarded-connect-retries) and use 0 as
417 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
418 * Set random(3) seed.
420 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
421 * Follow the OpenBSD port's lead and protect the resolve
422 * functions on OpenBSD as well.
424 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
425 * --user fix. Only use the user as group if no group is specified.
426 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
428 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
429 * Windows service (re-)integration
433 * --install[:service_name]
434 * --uninstall[:service_name]
437 * They work as follows:
438 * --install will create a service for you and then terminate.
439 * By default the service name will be "privoxy" (without the quotes).
440 * However you can run multiple services if you wish, just by adding
441 * a colon and then a name (no spaces).
443 * --uninstall follows the exact same rules a --install.
445 * --service is used when the program is executed by the service
446 * control manager, and in normal circumstances would never be
447 * used as a command line argument.
449 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
450 * Windows service integration
452 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
453 * Incorporate Fabian Keil's patch work:
454 http://www.fabiankeil.de/sourcecode/privoxy/
456 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
457 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
458 * with what was really the latest development (the v_3_0_branch branch)
460 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
461 * Thanks to Jindrich Makovicka for a race condition fix for the log
462 * file. The race condition remains for non-pthread implementations.
463 * Reference patch #1175720.
465 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
466 * Add the ability to check jpeg images for invalid
467 * lengths of comment blocks. Defensive strategy
468 * against the exploit:
469 * Microsoft Security Bulletin MS04-028
470 * Buffer Overrun in JPEG Processing (GDI+) Could
471 * Allow Code Execution (833987)
472 * Enabled with +inspect-jpegs in actions files.
474 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
475 * - Fixed usage info for non-unix platforms
476 * - Fixed small cmdline parsing bug
478 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
479 * Diagnostics: Now preserve the returncode of pthread_create
480 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
482 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
483 * No longer ignore SIGCHLD. Fixes bug #769381
485 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
486 * Close jarfile on graceful exit, remove stray line
488 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
489 * Cosmetics: Killed a warning, a typo and an allocation left at exit
491 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
492 * No longer rely on non-POSIX.1 extensions of getcwd().
495 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
496 * Replaced setenv() by posix-compliant putenv()
497 * Thanks to Neil McCalden (nmcc AT users.sf.net).
499 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
500 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
502 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
503 * Clean-up and extension of improvements for forked mode:
504 * - Child's return code now consists of flags RC_FLAG_*
505 * - Reporting toggle to parent now properly #ifdef'ed
506 * - Children now report blocking to parent. This enables
507 * statistics in forked mode
509 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
510 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
511 * Children now report having been toggled through _exit(17), parents
512 * watch for that code and toggle themselves if found.
514 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
515 * Wrapping all *_r functions (the non-_r versions of them) with
516 * mutex semaphores for OSX. Hopefully this will take care of all
517 * of those pesky crash reports.
519 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
520 * Fixed two mostly harmless mem leaks
522 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
523 * Fix: Head of global clients list now initialized to NULL
525 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
526 * Added basic support for OPTIONS and TRACE HTTP methods:
527 * - New interceptor direct_response() added in chat().
528 * - sed() moved to earlier in the process, so that the
529 * Host: header is evaluated before actions and forwarding
532 * Revision 1.92 2002/05/08 16:00:46 oes
533 * Chat's buffer handling:
534 * - Fixed bug with unchecked out-of-mem conditions
535 * while reading client request & server headers
536 * - No longer predict if the buffer limit will be exceeded
537 * in the next read -- check add_to_iob's new
538 * return code. If buffer couldn't be extended
539 * (policy or out-of-mem) while
540 * - reading from client: abort
541 * - reading server headers: send error page
542 * - buffering server body for filter: flush,
543 * and if that fails: send error page
545 * Revision 1.91 2002/04/08 20:35:58 swa
548 * Revision 1.90 2002/04/02 14:57:28 oes
549 * Made sending wafers independent of FEATURE_COOKIE_JAR
551 * Revision 1.89 2002/03/31 17:18:59 jongfoster
552 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
554 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
555 * More compiler warning message maintenance
557 * Revision 1.87 2002/03/26 22:29:54 swa
558 * we have a new homepage!
560 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
561 * Workaround for closing the jarfile before load_config() comes around again
563 * Revision 1.85 2002/03/24 15:23:33 jongfoster
566 * Revision 1.84 2002/03/24 13:25:43 swa
567 * name change related issues
569 * Revision 1.83 2002/03/16 23:54:06 jongfoster
570 * Adding graceful termination feature, to help look for memory leaks.
571 * If you enable this (which, by design, has to be done by hand
572 * editing config.h) and then go to http://i.j.b/die, then the program
573 * will exit cleanly after the *next* request. It should free all the
574 * memory that was used.
576 * Revision 1.82 2002/03/13 00:27:05 jongfoster
579 * Revision 1.81 2002/03/12 01:42:50 oes
580 * Introduced modular filters
582 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
583 * OS/2 port maintenance:
584 * - Fixed EMX build - it had decayed a little
585 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
586 * substituted a memset for now.
588 * Revision 1.79 2002/03/09 20:03:52 jongfoster
589 * - Making various functions return int rather than size_t.
590 * (Undoing a recent change). Since size_t is unsigned on
591 * Windows, functions like read_socket that return -1 on
592 * error cannot return a size_t.
594 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
595 * crashes, and also frequently caused JB to jump to 100%
596 * CPU and stay there. (Because it thought it had just
597 * read ((unsigned)-1) == 4Gb of data...)
599 * - The signature of write_socket has changed, it now simply
600 * returns success=0/failure=nonzero.
602 * - Trying to get rid of a few warnings --with-debug on
603 * Windows, I've introduced a new type "jb_socket". This is
604 * used for the socket file descriptors. On Windows, this
605 * is SOCKET (a typedef for unsigned). Everywhere else, it's
606 * an int. The error value can't be -1 any more, so it's
607 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
608 * Windows it maps to the #define INVALID_SOCKET.)
610 * - The signature of bind_port has changed.
612 * Revision 1.78 2002/03/08 21:35:04 oes
613 * Added optional group supplement to --user option. Will now use default group of user if no group given
615 * Revision 1.77 2002/03/07 03:52:06 oes
616 * - Fixed compiler warnings etc
617 * - Improved handling of failed DNS lookups
619 * Revision 1.76 2002/03/06 22:54:35 jongfoster
620 * Automated function-comment nitpicking.
622 * Revision 1.75 2002/03/06 10:02:19 oes
623 * Fixed stupid bug when --user was not given
625 * Revision 1.74 2002/03/06 00:49:31 jongfoster
626 * Fixing warning on Windows
627 * Making #ifdefs that refer to the same variable consistently
628 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
630 * Revision 1.73 2002/03/05 23:57:30 hal9
631 * Stray character 's' on line 1618 was breaking build.
633 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
634 * - Re-enable OS/2 building after new parms were added
635 * - Fix false out of memory report when resolving CGI templates when no IP
636 * address is available of failed attempt (a la no such domain)
638 * Revision 1.71 2002/03/05 18:13:56 oes
639 * Added --user option
641 * Revision 1.70 2002/03/05 04:52:42 oes
642 * Deleted non-errlog debugging code
644 * Revision 1.69 2002/03/04 23:50:00 jongfoster
645 * Splitting off bind_port() call into bind_port_helper(), with
648 * Revision 1.68 2002/03/04 20:17:32 oes
651 * Revision 1.67 2002/03/04 18:18:57 oes
652 * - Removed _DEBUG mode
653 * - Cleand up cmdline parsing
654 * - Introduced --no-daemon, --pidfile options
655 * - Cleaned up signal handling:
656 * - Terminate cleanly on INT, TERM and ABRT
657 * - Schedule logfile for re-opening on HUP
658 * - Ignore CHLD and PIPE
659 * - Leave the rest with their default handlers
660 * - Uniform handler registration
661 * - Added usage() function
662 * - Played styleguide police
664 * Revision 1.66 2002/03/03 15:06:55 oes
665 * Re-enabled automatic config reloading
667 * Revision 1.65 2002/03/03 14:49:11 oes
668 * Fixed CLF logging: Now uses client's original HTTP request
670 * Revision 1.64 2002/03/03 09:18:03 joergs
671 * Made jumbjuster work on AmigaOS again.
673 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
674 * Clean up a little CRLF unpleasantness that suddenly appeared
676 * Revision 1.62 2002/02/20 23:17:23 jongfoster
677 * Detecting some out-of memory conditions and exiting with a log message.
679 * Revision 1.61 2002/01/17 21:01:52 jongfoster
680 * Moving all our URL and URL pattern parsing code to urlmatch.c.
682 * Revision 1.60 2001/12/30 14:07:32 steudten
683 * - Add signal handling (unix)
684 * - Add SIGHUP handler (unix)
685 * - Add creation of pidfile (unix)
686 * - Add action 'top' in rc file (RH)
687 * - Add entry 'SIGNALS' to manpage
688 * - Add exit message to logfile (unix)
690 * Revision 1.59 2001/12/13 14:07:18 oes
691 * Fixed Bug: 503 error page now sent OK
693 * Revision 1.58 2001/11/30 23:37:24 jongfoster
694 * Renaming the Win32 config file to config.txt - this is almost the
695 * same as the corresponding UNIX name "config"
697 * Revision 1.57 2001/11/16 00:47:43 jongfoster
698 * Changing the tty-disconnection code to use setsid().
700 * Revision 1.56 2001/11/13 20:20:54 jongfoster
701 * Tabs->spaces, fixing a bug with missing {} around an if()
703 * Revision 1.55 2001/11/13 20:14:53 jongfoster
704 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
706 * Revision 1.54 2001/11/07 00:03:14 steudten
707 * Give reliable return value if an error
708 * occurs not just 0 with new daemon mode.
710 * Revision 1.53 2001/11/05 21:41:43 steudten
711 * Add changes to be a real daemon just for unix os.
712 * (change cwd to /, detach from controlling tty, set
713 * process group and session leader to the own process.
715 * Add some fatal-error log message for failed malloc().
716 * Add '-d' if compiled with 'configure --with-debug' to
717 * enable debug output.
719 * Revision 1.52 2001/10/26 20:11:20 jongfoster
720 * Fixing type mismatch
722 * Revision 1.51 2001/10/26 17:38:28 oes
725 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
726 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
727 * threads to call select() simultaneously. So, it's time to do a real, live,
728 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
729 * (native). Both versions will work, but using __OS2__ offers multi-threading.
731 * Revision 1.49 2001/10/23 21:41:35 jongfoster
732 * Added call to initialize the (statically-allocated of course)
733 * "out of memory" CGI response.
735 * Revision 1.48 2001/10/10 19:56:46 jongfoster
736 * Moving some code that wasn't cookie-related out of an #ifdef
739 * Revision 1.47 2001/10/10 16:44:36 oes
740 * Added CONNECT destination port limitation check
742 * Revision 1.46 2001/10/08 15:17:41 oes
743 * Re-enabled SSL forwarding
745 * Revision 1.45 2001/10/07 15:42:11 oes
746 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
748 * Moved downgrading of the HTTP version from parse_http_request to
749 * chat(), since we can't decide if it is necessary before we have
750 * determined the actions for the URL. The HTTP command is now
751 * *always* re-built so the repairs need no longer be special-cased.
753 * filter_popups now gets a csp pointer so it can raise the new
754 * CSP_FLAG_MODIFIED flag.
758 * Added configurable size limit for the IOB. If the IOB grows so
759 * large that the next read would exceed the limit, the header
760 * is generated, and the header & unfiltered buffer are flushed
761 * to the client. Chat then continues in non-buffering,
762 * non-filtering body mode.
764 * Revision 1.44 2001/10/02 18:13:57 oes
767 * Revision 1.43 2001/10/02 15:32:13 oes
768 * Moved generation of hdr
770 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
771 * Cleaning up 2 compiler warnings on OS/2.
773 * Revision 1.41 2001/09/16 17:05:14 jongfoster
774 * Removing unused #include showarg.h
776 * Revision 1.40 2001/09/16 15:41:45 jongfoster
777 * Fixing signed/unsigned comparison warning.
779 * Revision 1.39 2001/09/16 13:21:27 jongfoster
780 * Changes to use new list functions.
782 * Revision 1.38 2001/09/16 13:01:46 jongfoster
783 * Removing redundant function call that zeroed zalloc()'d memory.
785 * Revision 1.37 2001/09/10 11:12:24 oes
786 * Deleted unused variable
788 * Revision 1.36 2001/09/10 10:56:15 oes
789 * Silenced compiler warnings
791 * Revision 1.35 2001/07/31 14:44:22 oes
792 * Deleted unused size parameter from filter_popups()
794 * Revision 1.34 2001/07/30 22:08:36 jongfoster
795 * Tidying up #defines:
796 * - All feature #defines are now of the form FEATURE_xxx
797 * - Permanently turned off WIN_GUI_EDIT
798 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
800 * Revision 1.33 2001/07/29 19:32:00 jongfoster
801 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
803 * Revision 1.32 2001/07/29 18:47:05 jongfoster
804 * Adding missing #include "loadcfg.h"
806 * Revision 1.31 2001/07/29 12:17:48 oes
807 * Applied pthread fix by Paul Lieverse
809 * Revision 1.30 2001/07/25 22:57:13 jongfoster
810 * __BEOS__ no longer overrides FEATURE_PTHREAD.
811 * This is because FEATURE_PTHREAD will soon be widely used, so I
812 * want to keep it simple.
814 * Revision 1.29 2001/07/24 12:47:06 oes
815 * Applied BeOS support update by Eugenia
817 * Revision 1.28 2001/07/23 13:26:12 oes
818 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
820 * Revision 1.27 2001/07/19 19:09:47 haroon
821 * - Added code to take care of the situation where while processing the first
822 * server response (which includes the server header), after finding the end
823 * of the headers we were not looking past the end of the headers for
824 * content modification. I enabled it for filter_popups.
825 * Someone else should look to see if other similar operations should be
826 * done to the discarded portion of the buffer.
828 * Note 2001/07/20: No, the other content modification mechanisms will process
829 * the whole iob later anyway. --oes
831 * Revision 1.26 2001/07/18 12:31:36 oes
834 * Revision 1.25 2001/07/15 19:43:49 jongfoster
835 * Supports POSIX threads.
836 * Also removed some unused #includes.
838 * Revision 1.24 2001/07/13 14:00:40 oes
839 * - Generic content modification scheme:
840 * Each feature has its own applicability flag that is set
841 * from csp->action->flags.
842 * Replaced the "filtering" int flag , by a function pointer
843 * "content_filter" to the function that will do the content
844 * modification. If it is != NULL, the document will be buffered
845 * and processed through *content_filter, which must set
846 * csp->content_length and return a modified copy of the body
847 * or return NULL (on failiure).
848 * - Changed csp->is_text to the more generic bitmap csp->content_type
849 * which can currently take the valued CT_TEXT or CT_GIF
851 * - Removed all #ifdef PCRS
853 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
854 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
855 * the `block_acl' call. This prevents a compilation error when the user
856 * does not wish to use the "ACL" feature.
858 * Revision 1.22 2001/06/29 21:45:41 oes
859 * Indentation, CRLF->LF, Tab-> Space
861 * Revision 1.21 2001/06/29 13:29:36 oes
862 * - Cleaned up, improved comments
863 * - Unified all possible interceptors (CGI,
864 * block, trust, fast_redirect) in one
865 * place, with one (CGI) answer generation
866 * mechansim. Much clearer now.
867 * - Removed the GIF image generation, which
868 * is now done in filters.c:block_url()
869 * - Made error conditions like domain lookup
870 * failiure or (various) problems while talking
871 * to the server use cgi.c:error_response()
872 * instead of generating HTML/HTTP in chat() (yuck!)
873 * - Removed logentry from cancelled commit
875 * Revision 1.20 2001/06/09 10:55:28 jongfoster
876 * Changing BUFSIZ ==> BUFFER_SIZE
878 * Revision 1.19 2001/06/07 23:12:52 jongfoster
879 * Replacing function pointer in struct gateway with a directly
880 * called function forwarded_connect().
881 * Replacing struct gateway with struct forward_spec
883 * Revision 1.18 2001/06/03 19:12:16 oes
884 * introduced new cgi handling
886 * Revision 1.17 2001/06/01 20:07:23 jongfoster
887 * Now uses action +image-blocker{} rather than config->tinygif
889 * Revision 1.16 2001/06/01 18:49:17 jongfoster
890 * Replaced "list_share" with "list" - the tiny memory gain was not
891 * worth the extra complexity.
893 * Revision 1.15 2001/05/31 21:24:47 jongfoster
894 * Changed "permission" to "action" throughout.
895 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
896 * Moved vanilla wafer check into chat(), since we must now
897 * decide whether or not to add it based on the URL.
899 * Revision 1.14 2001/05/29 20:14:01 joergs
900 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
903 * Revision 1.13 2001/05/29 09:50:24 jongfoster
904 * Unified blocklist/imagelist/permissionslist.
905 * File format is still under discussion, but the internal changes
908 * Also modified interceptor behaviour:
909 * - We now intercept all URLs beginning with one of the following
910 * prefixes (and *only* these prefixes):
912 * * http://ijbswa.sf.net/config/
913 * * http://ijbswa.sourceforge.net/config/
914 * - New interceptors "home page" - go to http://i.j.b/ to see it.
915 * - Internal changes so that intercepted and fast redirect pages
916 * are not replaced with an image.
917 * - Interceptors now have the option to send a binary page direct
918 * to the client. (i.e. ijb-send-banner uses this)
919 * - Implemented show-url-info interceptor. (Which is why I needed
920 * the above interceptors changes - a typical URL is
921 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
922 * The previous mechanism would not have intercepted that, and
923 * if it had been intercepted then it then it would have replaced
926 * Revision 1.12 2001/05/27 22:17:04 oes
928 * - re_process_buffer no longer writes the modified buffer
929 * to the client, which was very ugly. It now returns the
930 * buffer, which it is then written by chat.
932 * - content_length now adjusts the Content-Length: header
933 * for modified documents rather than crunch()ing it.
934 * (Length info in csp->content_length, which is 0 for
935 * unmodified documents)
937 * - For this to work, sed() is called twice when filtering.
939 * Revision 1.11 2001/05/26 17:27:53 jongfoster
940 * Added support for CLF and fixed LOG_LEVEL_LOG.
941 * Also did CRLF->LF fix of my previous patch.
943 * Revision 1.10 2001/05/26 15:26:15 jongfoster
944 * ACL feature now provides more security by immediately dropping
945 * connections from untrusted hosts.
947 * Revision 1.9 2001/05/26 00:28:36 jongfoster
948 * Automatic reloading of config file.
949 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
950 * Most of the global variables have been moved to a new
951 * struct configuration_spec, accessed through csp->config->globalname
952 * Most of the globals remaining are used by the Win32 GUI.
954 * Revision 1.8 2001/05/25 22:43:18 jongfoster
955 * Fixing minor memory leak and buffer overflow.
957 * Revision 1.7 2001/05/25 22:34:30 jongfoster
960 * Revision 1.6 2001/05/23 00:13:58 joergs
961 * AmigaOS support fixed.
963 * Revision 1.5 2001/05/22 18:46:04 oes
965 * - Enabled filtering banners by size rather than URL
966 * by adding patterns that replace all standard banner
967 * sizes with the "Junkbuster" gif to the re_filterfile
969 * - Enabled filtering WebBugs by providing a pattern
970 * which kills all 1x1 images
972 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
973 * which is selected by the (nonstandard and therefore
974 * capital) letter 'U' in the option string.
975 * It causes the quantifiers to be ungreedy by default.
976 * Appending a ? turns back to greedy (!).
978 * - Added a new interceptor ijb-send-banner, which
979 * sends back the "Junkbuster" gif. Without imagelist or
980 * MSIE detection support, or if tinygif = 1, or the
981 * URL isn't recognized as an imageurl, a lame HTML
982 * explanation is sent instead.
984 * - Added new feature, which permits blocking remote
985 * script redirects and firing back a local redirect
987 * The feature is conditionally compiled, i.e. it
988 * can be disabled with --disable-fast-redirects,
989 * plus it must be activated by a "fast-redirects"
990 * line in the config file, has its own log level
991 * and of course wants to be displayed by show-proxy-args
992 * Note: Boy, all the #ifdefs in 1001 locations and
993 * all the fumbling with configure.in and acconfig.h
994 * were *way* more work than the feature itself :-(
996 * - Because a generic redirect template was needed for
997 * this, tinygif = 3 now uses the same.
999 * - Moved GIFs, and other static HTTP response templates
1002 * - Some minor fixes
1004 * - Removed some >400 CRs again (Jon, you really worked
1007 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1008 * Made failure to bind() a fatal error.
1010 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1011 * Version 2.9.4 checkin.
1012 * - Merged popupfile and cookiefile, and added control over PCRS
1013 * filtering, in new "permissionsfile".
1014 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1015 * file error you now get a message box (in the Win32 GUI) rather
1016 * than the program exiting with no explanation.
1017 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1019 * - Removed tabs from "config"
1020 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1021 * - Bumped up version number.
1023 * Revision 1.2 2001/05/17 22:34:44 oes
1024 * - Added hint on GIF char array generation to jcc.c
1025 * - Cleaned CRLF's from the sources and related files
1026 * - Repaired logging for REF and FRC
1028 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1029 * Initial import of version 2.9.3 source tree
1032 *********************************************************************/
1038 #include <sys/types.h>
1047 # ifndef FEATURE_PTHREAD
1051 # include <windows.h>
1052 # include <process.h>
1053 # endif /* ndef FEATURE_PTHREAD */
1056 # ifndef _WIN_CONSOLE
1057 # include "w32log.h"
1058 # endif /* ndef _WIN_CONSOLE */
1059 # include "w32svrapi.h"
1061 #else /* ifndef _WIN32 */
1063 # if !defined (__OS2__)
1064 # include <unistd.h>
1065 # include <sys/wait.h>
1066 # endif /* ndef __OS2__ */
1067 # include <sys/time.h>
1068 # include <sys/stat.h>
1069 # include <sys/ioctl.h>
1072 #include <sys/termios.h>
1080 # include <signal.h>
1083 # include <socket.h> /* BeOS has select() for sockets only. */
1084 # include <OS.h> /* declarations for threads and stuff. */
1087 # if defined(__EMX__) || defined(__OS2__)
1088 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1093 #define bzero(B,N) memset(B,0x00,n)
1097 # include <select.h>
1102 #include "project.h"
1105 #include "filters.h"
1106 #include "loaders.h"
1107 #include "parsers.h"
1108 #include "miscutil.h"
1110 #include "jbsockets.h"
1111 #include "gateway.h"
1112 #include "actions.h"
1114 #include "loadcfg.h"
1115 #include "urlmatch.h"
1117 const char jcc_h_rcs[] = JCC_H_VERSION;
1118 const char project_h_rcs[] = PROJECT_H_VERSION;
1121 struct client_state clients[1];
1122 struct file_list files[1];
1124 #ifdef FEATURE_STATISTICS
1125 int urls_read = 0; /* total nr of urls read inc rejected */
1126 int urls_rejected = 0; /* total nr of urls rejected */
1127 #endif /* def FEATURE_STATISTICS */
1129 #ifdef FEATURE_GRACEFUL_TERMINATION
1130 int g_terminate = 0;
1133 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1134 static void sig_handler(int the_signal);
1136 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1137 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1138 static jb_err get_server_headers(struct client_state *csp);
1139 static const char *crunch_reason(const struct http_response *rsp);
1140 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1142 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1144 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1145 static jb_err change_request_destination(struct client_state *csp);
1146 static void chat(struct client_state *csp);
1147 static void serve(struct client_state *csp);
1148 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1149 static void usage(const char *myname);
1151 static void initialize_mutexes(void);
1152 static jb_socket bind_port_helper(struct configuration_spec *config);
1153 static void listen_loop(void);
1156 void serve(struct client_state *csp);
1157 #else /* ifndef AMIGA */
1158 static void serve(struct client_state *csp);
1159 #endif /* def AMIGA */
1162 static int32 server_thread(void *data);
1163 #endif /* def __BEOS__ */
1166 #define sleep(N) Sleep(((N) * 1000))
1170 #define sleep(N) DosSleep(((N) * 100))
1173 #ifdef FEATURE_PTHREAD
1175 * XXX: Does the locking stuff really belong in this file?
1177 privoxy_mutex_t log_mutex;
1178 privoxy_mutex_t log_init_mutex;
1180 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1181 privoxy_mutex_t resolver_mutex;
1182 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1184 #ifndef HAVE_GMTIME_R
1185 privoxy_mutex_t gmtime_mutex;
1186 #endif /* ndef HAVE_GMTIME_R */
1188 #ifndef HAVE_LOCALTIME_R
1189 privoxy_mutex_t localtime_mutex;
1190 #endif /* ndef HAVE_GMTIME_R */
1193 privoxy_mutex_t rand_mutex;
1194 #endif /* ndef HAVE_RANDOM */
1196 #endif /* FEATURE_PTHREAD */
1199 const char *basedir = NULL;
1200 const char *pidfile = NULL;
1201 static int received_hup_signal = 0;
1202 #endif /* defined unix */
1204 /* HTTP snipplets. */
1205 static const char CSUCCEED[] =
1206 "HTTP/1.0 200 Connection established\r\n"
1207 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1209 static const char CHEADER[] =
1210 "HTTP/1.0 400 Invalid header received from client\r\n"
1211 "Proxy-Agent: Privoxy " VERSION "\r\n"
1212 "Content-Type: text/plain\r\n"
1213 "Connection: close\r\n\r\n"
1214 "Invalid header received from client.\r\n";
1216 static const char FTP_RESPONSE[] =
1217 "HTTP/1.0 400 Invalid request received from client\r\n"
1218 "Content-Type: text/plain\r\n"
1219 "Connection: close\r\n\r\n"
1220 "Invalid request. Privoxy doesn't support FTP.\r\n";
1222 static const char GOPHER_RESPONSE[] =
1223 "HTTP/1.0 400 Invalid request received from client\r\n"
1224 "Content-Type: text/plain\r\n"
1225 "Connection: close\r\n\r\n"
1226 "Invalid request. Privoxy doesn't support gopher.\r\n";
1228 /* XXX: should be a template */
1229 static const char MISSING_DESTINATION_RESPONSE[] =
1230 "HTTP/1.0 400 Bad request received from client\r\n"
1231 "Proxy-Agent: Privoxy " VERSION "\r\n"
1232 "Content-Type: text/plain\r\n"
1233 "Connection: close\r\n\r\n"
1234 "Bad request. Privoxy was unable to extract the destination.\r\n";
1236 /* XXX: should be a template */
1237 static const char NO_SERVER_DATA_RESPONSE[] =
1238 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1239 "Proxy-Agent: Privoxy " VERSION "\r\n"
1240 "Content-Type: text/plain\r\n"
1241 "Connection: close\r\n\r\n"
1242 "Empty server or forwarder response.\r\n"
1243 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1245 /* XXX: should be a template */
1246 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1247 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1248 "Proxy-Agent: Privoxy " VERSION "\r\n"
1249 "Content-Type: text/plain\r\n"
1250 "Connection: close\r\n\r\n"
1251 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1254 /* XXX: should be a template */
1255 static const char NULL_BYTE_RESPONSE[] =
1256 "HTTP/1.0 400 Bad request received from client\r\n"
1257 "Proxy-Agent: Privoxy " VERSION "\r\n"
1258 "Content-Type: text/plain\r\n"
1259 "Connection: close\r\n\r\n"
1260 "Bad request. Null byte(s) before end of request.\r\n";
1263 /* XXX: should be a template */
1264 static const char MESSED_UP_REQUEST_RESPONSE[] =
1265 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1266 "Proxy-Agent: Privoxy " VERSION "\r\n"
1267 "Content-Type: text/plain\r\n"
1268 "Connection: close\r\n\r\n"
1269 "Bad request. Messed up with header filters.\r\n";
1271 /* A function to crunch a response */
1272 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1274 /* Crunch function flags */
1275 #define CF_NO_FLAGS 0
1276 /* Cruncher applies to forced requests as well */
1277 #define CF_IGNORE_FORCE 1
1278 /* Crunched requests are counted for the block statistics */
1279 #define CF_COUNT_AS_REJECT 2
1281 /* A crunch function and its flags */
1284 const crunch_func_ptr cruncher;
1288 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1290 /* Complete list of cruncher functions */
1291 static const struct cruncher crunchers_all[] = {
1292 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1293 { block_url, CF_COUNT_AS_REJECT },
1294 #ifdef FEATURE_TRUST
1295 { trust_url, CF_COUNT_AS_REJECT },
1296 #endif /* def FEATURE_TRUST */
1297 { redirect_url, CF_NO_FLAGS },
1298 { dispatch_cgi, CF_IGNORE_FORCE},
1302 /* Light version, used after tags are applied */
1303 static const struct cruncher crunchers_light[] = {
1304 { block_url, CF_COUNT_AS_REJECT },
1305 { redirect_url, CF_NO_FLAGS },
1311 * XXX: Don't we really mean
1317 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1318 /*********************************************************************
1320 * Function : sig_handler
1322 * Description : Signal handler for different signals.
1323 * Exit gracefully on TERM and INT
1324 * or set a flag that will cause the errlog
1325 * to be reopened by the main thread on HUP.
1328 * 1 : the_signal = the signal cause this function to call
1332 *********************************************************************/
1333 static void sig_handler(int the_signal)
1339 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1351 received_hup_signal = 1;
1357 * We shouldn't be here, unless we catch signals
1358 * in main() that we can't handle here!
1360 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1368 /*********************************************************************
1370 * Function : client_protocol_is_unsupported
1372 * Description : Checks if the client used a known unsupported
1373 * protocol and deals with it by sending an error
1377 * 1 : csp = Current client state (buffers, headers, etc...)
1378 * 2 : req = the first request line send by the client
1380 * Returns : TRUE if an error response has been generated, or
1381 * FALSE if the request doesn't look invalid.
1383 *********************************************************************/
1384 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1387 * If it's a FTP or gopher request, we don't support it.
1389 * These checks are better than nothing, but they might
1390 * not work in all configurations and some clients might
1391 * have problems digesting the answer.
1393 * They should, however, never cause more problems than
1394 * Privoxy's old behaviour (returning the misleading HTML
1397 * "Could not resolve http://(ftp|gopher)://example.org").
1399 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1401 const char *response = NULL;
1402 const char *protocol = NULL;
1404 if (!strncmpic(req, "GET ftp://", 10))
1406 response = FTP_RESPONSE;
1411 response = GOPHER_RESPONSE;
1412 protocol = "GOPHER";
1414 log_error(LOG_LEVEL_ERROR,
1415 "%s tried to use Privoxy as %s proxy: %s",
1416 csp->ip_addr_str, protocol, req);
1417 log_error(LOG_LEVEL_CLF,
1418 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1420 write_socket(csp->cfd, response, strlen(response));
1429 /*********************************************************************
1431 * Function : get_request_destination_elsewhere
1433 * Description : If the client's request was redirected into
1434 * Privoxy without the client's knowledge,
1435 * the request line lacks the destination host.
1437 * This function tries to get it elsewhere,
1438 * provided accept-intercepted-requests is enabled.
1440 * "Elsewhere" currently only means "Host: header",
1441 * but in the future we may ask the redirecting
1442 * packet filter to look the destination up.
1444 * If the destination stays unknown, an error
1445 * response is send to the client and headers
1446 * are freed so that chat() can return directly.
1449 * 1 : csp = Current client state (buffers, headers, etc...)
1450 * 2 : headers = a header list
1452 * Returns : JB_ERR_OK if the destination is now known, or
1453 * JB_ERR_PARSE if it isn't.
1455 *********************************************************************/
1456 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1460 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1462 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1463 " Privoxy isn't configured to accept intercepted requests.",
1464 csp->ip_addr_str, csp->http->cmd);
1465 /* XXX: Use correct size */
1466 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1467 csp->ip_addr_str, csp->http->cmd);
1469 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1470 destroy_list(headers);
1472 return JB_ERR_PARSE;
1474 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1476 /* Split the domain we just got for pattern matching */
1477 init_domain_components(csp->http);
1483 /* We can't work without destination. Go spread the news.*/
1485 req = list_to_text(headers);
1487 /* XXX: Use correct size */
1488 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1489 csp->ip_addr_str, csp->http->cmd);
1490 log_error(LOG_LEVEL_ERROR,
1491 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1492 csp->ip_addr_str, csp->http->cmd, req);
1495 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1496 destroy_list(headers);
1498 return JB_ERR_PARSE;
1501 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1502 * to get the destination IP address, use it as host directly
1503 * or do a reverse DNS lookup first.
1508 /*********************************************************************
1510 * Function : get_server_headers
1512 * Description : Parses server headers in iob and fills them
1513 * into csp->headers so that they can later be
1517 * 1 : csp = Current client state (buffers, headers, etc...)
1519 * Returns : JB_ERR_OK if everything went fine, or
1520 * JB_ERR_PARSE if the headers were incomplete.
1522 *********************************************************************/
1523 static jb_err get_server_headers(struct client_state *csp)
1525 int continue_hack_in_da_house = 0;
1528 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1533 * continue hack in da house. Ignore the ending of
1534 * this head and continue enlisting header lines.
1535 * The reason is described below.
1537 enlist(csp->headers, "");
1538 continue_hack_in_da_house = 0;
1541 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1544 * It's a bodyless continue response, don't
1545 * stop header parsing after reaching its end.
1547 * As a result Privoxy will concatenate the
1548 * next response's head and parse and deliver
1549 * the headers as if they belonged to one request.
1551 * The client will separate them because of the
1552 * empty line between them.
1554 * XXX: What we're doing here is clearly against
1555 * the intended purpose of the continue header,
1556 * and under some conditions (HTTP/1.0 client request)
1557 * it's a standard violation.
1559 * Anyway, "sort of against the spec" is preferable
1560 * to "always getting confused by Continue responses"
1561 * (Privoxy's behaviour before this hack was added)
1563 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1564 continue_hack_in_da_house = 1;
1566 else if (*header == '\0')
1569 * If the header is empty, but the Continue hack
1570 * isn't active, we can assume that we reached the
1571 * end of the buffer before we hit the end of the
1574 * Inform the caller an let it decide how to handle it.
1576 return JB_ERR_PARSE;
1579 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1582 * XXX: Should we quit the request and return a
1583 * out of memory error page instead?
1585 log_error(LOG_LEVEL_ERROR,
1586 "Out of memory while enlisting server headers. %s lost.",
1596 /*********************************************************************
1598 * Function : crunch_reason
1600 * Description : Translates the crunch reason code into a string.
1603 * 1 : rsp = a http_response
1605 * Returns : A string with the crunch reason or an error description.
1607 *********************************************************************/
1608 static const char *crunch_reason(const struct http_response *rsp)
1610 char * reason = NULL;
1612 assert(rsp != NULL);
1615 return "Internal error while searching for crunch reason";
1618 switch (rsp->reason)
1620 case RSP_REASON_UNSUPPORTED:
1621 reason = "Unsupported HTTP feature";
1623 case RSP_REASON_BLOCKED:
1626 case RSP_REASON_UNTRUSTED:
1627 reason = "Untrusted";
1629 case RSP_REASON_REDIRECTED:
1630 reason = "Redirected";
1632 case RSP_REASON_CGI_CALL:
1633 reason = "CGI Call";
1635 case RSP_REASON_NO_SUCH_DOMAIN:
1636 reason = "DNS failure";
1638 case RSP_REASON_FORWARDING_FAILED:
1639 reason = "Forwarding failed";
1641 case RSP_REASON_CONNECT_FAILED:
1642 reason = "Connection failure";
1644 case RSP_REASON_OUT_OF_MEMORY:
1645 reason = "Out of memory (may mask other reasons)";
1648 reason = "No reason recorded";
1656 /*********************************************************************
1658 * Function : send_crunch_response
1660 * Description : Delivers already prepared response for
1661 * intercepted requests, logs the interception
1662 * and frees the response.
1665 * 1 : csp = Current client state (buffers, headers, etc...)
1666 * 1 : rsp = Fully prepared response. Will be freed on exit.
1668 * Returns : Nothing.
1670 *********************************************************************/
1671 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1673 const struct http_request *http = csp->http;
1674 char status_code[4];
1676 assert(rsp != NULL);
1677 assert(rsp->head != NULL);
1682 * Not supposed to happen. If it does
1683 * anyway, treat it as an unknown error.
1685 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1686 /* return code doesn't matter */
1691 /* If rsp is still NULL, we have serious internal problems. */
1692 log_error(LOG_LEVEL_FATAL,
1693 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1697 * Extract the status code from the actual head
1698 * that was send to the client. It is the only
1699 * way to get it right for all requests, including
1700 * the fixed ones for out-of-memory problems.
1702 * A head starts like this: 'HTTP/1.1 200...'
1706 status_code[0] = rsp->head[9];
1707 status_code[1] = rsp->head[10];
1708 status_code[2] = rsp->head[11];
1709 status_code[3] = '\0';
1711 /* Write the answer to the client */
1712 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1713 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1715 /* There is nothing we can do about it. */
1716 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1719 /* Log that the request was crunched and why. */
1720 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1721 http->hostport, http->path, crunch_reason(rsp));
1722 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1723 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1725 /* Clean up and return */
1726 if (cgi_error_memory() != rsp)
1728 free_http_response(rsp);
1735 /*********************************************************************
1737 * Function : request_contains_null_bytes
1739 * Description : Checks for NULL bytes in the request and sends
1740 * an error message to the client if any were found.
1742 * XXX: currently not used, see comment in chat().
1745 * 1 : csp = Current client state (buffers, headers, etc...)
1746 * 2 : buf = Data from the client's request to check.
1747 * 3 : len = The data length.
1749 * Returns : TRUE if the request contained one or more NULL bytes, or
1752 *********************************************************************/
1753 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1755 size_t c_len; /* Request lenght when treated as C string */
1757 c_len = strlen(buf);
1762 * Null byte(s) found. Log the request,
1763 * return an error response and hang up.
1765 size_t tmp_len = c_len;
1770 * Replace NULL byte(s) with '°' characters
1771 * so the request can be logged as string.
1772 * XXX: Is there a better replacement character?
1775 tmp_len += strlen(buf+tmp_len);
1776 } while (tmp_len < len);
1778 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1779 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1780 log_error(LOG_LEVEL_HEADER,
1781 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1783 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1785 /* XXX: Log correct size */
1786 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1796 /*********************************************************************
1798 * Function : crunch_response_triggered
1800 * Description : Checks if the request has to be crunched,
1801 * and delivers the crunch response if necessary.
1804 * 1 : csp = Current client state (buffers, headers, etc...)
1805 * 2 : crunchers = list of cruncher functions to run
1807 * Returns : TRUE if the request was answered with a crunch response
1810 *********************************************************************/
1811 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1813 struct http_response *rsp = NULL;
1814 const struct cruncher *c;
1817 * If CGI request crunching is disabled,
1818 * check the CGI dispatcher out of order to
1819 * prevent unintentional blocks or redirects.
1821 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1822 && (NULL != (rsp = dispatch_cgi(csp))))
1824 /* Deliver, log and free the interception response. */
1825 send_crunch_response(csp, rsp);
1829 for (c = crunchers; c->cruncher != NULL; c++)
1832 * Check the cruncher if either Privoxy is toggled
1833 * on and the request isn't forced, or if the cruncher
1834 * applies to forced requests as well.
1836 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1837 !(csp->flags & CSP_FLAG_FORCED)) ||
1838 (c->flags & CF_IGNORE_FORCE))
1840 rsp = c->cruncher(csp);
1843 /* Deliver, log and free the interception response. */
1844 send_crunch_response(csp, rsp);
1845 #ifdef FEATURE_STATISTICS
1846 if (c->flags & CF_COUNT_AS_REJECT)
1848 csp->flags |= CSP_FLAG_REJECTED;
1850 #endif /* def FEATURE_STATISTICS */
1861 /*********************************************************************
1863 * Function : build_request_line
1865 * Description : Builds the HTTP request line.
1867 * If a HTTP forwarder is used it expects the whole URL,
1868 * web servers only get the path.
1871 * 1 : csp = Current client state (buffers, headers, etc...)
1872 * 2 : fwd = The forwarding spec used for the request
1873 * XXX: Should use http->fwd instead.
1874 * 3 : request_line = The old request line which will be replaced.
1876 * Returns : Nothing. Terminates in case of memory problems.
1878 *********************************************************************/
1879 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1881 struct http_request *http = csp->http;
1883 assert(http->ssl == 0);
1886 * Downgrade http version from 1.1 to 1.0
1887 * if +downgrade action applies.
1889 if ( (csp->action->flags & ACTION_DOWNGRADE)
1890 && (!strcmpic(http->ver, "HTTP/1.1")))
1893 http->ver = strdup("HTTP/1.0");
1895 if (http->ver == NULL)
1897 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1902 * Rebuild the request line.
1904 freez(*request_line);
1905 *request_line = strdup(http->gpc);
1906 string_append(request_line, " ");
1908 if (fwd->forward_host)
1910 string_append(request_line, http->url);
1914 string_append(request_line, http->path);
1916 string_append(request_line, " ");
1917 string_append(request_line, http->ver);
1919 if (*request_line == NULL)
1921 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1923 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1927 /*********************************************************************
1929 * Function : change_request_destination
1931 * Description : Parse a (rewritten) request line and regenerate
1932 * the http request data.
1935 * 1 : csp = Current client state (buffers, headers, etc...)
1937 * Returns : Forwards the parse_http_request() return code.
1938 * Terminates in case of memory problems.
1940 *********************************************************************/
1941 static jb_err change_request_destination(struct client_state *csp)
1943 struct http_request *http = csp->http;
1946 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1947 free_http_request(http);
1948 err = parse_http_request(csp->headers->first->str, http, csp);
1949 if (JB_ERR_OK != err)
1951 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1952 jb_err_to_string(err));
1956 /* XXX: ocmd is a misleading name */
1957 http->ocmd = strdup(http->cmd);
1958 if (http->ocmd == NULL)
1960 log_error(LOG_LEVEL_FATAL,
1961 "Out of memory copying rewritten HTTP request line");
1969 /*********************************************************************
1973 * Description : Once a connection to the client has been accepted,
1974 * this function is called (via serve()) to handle the
1975 * main business of the communication. When this
1976 * function returns, the caller must close the client
1979 * FIXME: chat is nearly thousand lines long.
1983 * 1 : csp = Current client state (buffers, headers, etc...)
1985 * Returns : Nothing.
1987 *********************************************************************/
1988 static void chat(struct client_state *csp)
1990 char buf[BUFFER_SIZE];
1998 int ms_iis5_hack = 0;
1999 size_t byte_count = 0;
2000 int forwarded_connect_retries = 0;
2001 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2002 const struct forward_spec * fwd;
2003 struct http_request *http;
2004 int len; /* for buffer sizes (and negative error codes) */
2007 /* Function that does the content filtering for the current request */
2008 filter_function_ptr content_filter = NULL;
2010 /* Skeleton for HTTP response, if we should intercept the request */
2011 struct http_response *rsp;
2013 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2014 struct list header_list;
2015 struct list *headers = &header_list;
2019 memset(buf, 0, sizeof(buf));
2022 * Read the client's request. Note that since we're not using select() we
2023 * could get blocked here if a client connected, then didn't say anything!
2028 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2030 if (len <= 0) break; /* error! */
2033 * If there is no memory left for buffering the
2034 * request, there is nothing we can do but hang up
2036 if (add_to_iob(csp, buf, len))
2041 req = get_header(csp->iob);
2043 } while ((NULL != req) && ('\0' == *req));
2045 if ((NULL != req) && ('\0' != *req))
2047 /* Request received. Validate and parse it. */
2051 * XXX: Temporary disabled to prevent problems
2052 * with POST requests whose bodies are allowed to
2053 * contain NULL bytes. BR#1730105.
2055 * The main purpose of this check is to properly
2056 * log stuff like BitTorrent traffic and other junk
2057 * that hits public proxies. It's not required for
2058 * Privoxy to functions as those requests are discarded
2061 * It probably should be rewritten to only check
2062 * the head of the request. Another option would
2063 * be to let all POST requests pass, although that
2064 * may not be good enough.
2066 if (request_contains_null_bytes(csp, buf, len))
2068 /* NULL bytes found and dealt with, just hang up. */
2073 /* Does the request line look invalid? */
2074 if (client_protocol_is_unsupported(csp, req))
2077 * Yes. The request has already been
2078 * answered with a error response, the buffers
2079 * were freed and we're done with chatting.
2084 #ifdef FEATURE_FORCE_LOAD
2086 * If this request contains the FORCE_PREFIX and blocks
2087 * aren't enforced, get rid of it and set the force flag.
2089 if (strstr(req, FORCE_PREFIX))
2091 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2093 log_error(LOG_LEVEL_FORCE,
2094 "Ignored force prefix in request: \"%s\".", req);
2098 strclean(req, FORCE_PREFIX);
2099 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2100 csp->flags |= CSP_FLAG_FORCED;
2104 #endif /* def FEATURE_FORCE_LOAD */
2105 err = parse_http_request(req, http, csp);
2106 if (JB_ERR_OK != err)
2108 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2114 if (http->cmd == NULL)
2116 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2117 /* XXX: Use correct size */
2118 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2119 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2121 free_http_request(http);
2125 /* grab the rest of the client's headers */
2129 p = get_header(csp->iob);
2133 /* There are no additional headers to read. */
2140 * We didn't receive a complete header
2141 * line yet, get the rest of it.
2143 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2146 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2147 destroy_list(headers);
2151 if (add_to_iob(csp, buf, len))
2154 * If there is no memory left for buffering the
2155 * request, there is nothing we can do but hang up
2157 destroy_list(headers);
2164 * We were able to read a complete
2165 * header and can finaly enlist it.
2172 if (http->host == NULL)
2175 * If we still don't know the request destination,
2176 * the request is invalid or the client uses
2177 * Privoxy without its knowledge.
2179 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2182 * Our attempts to get the request destination
2183 * elsewhere failed or Privoxy is configured
2184 * to only accept proxy requests.
2186 * An error response has already been send
2187 * and we're done here.
2194 * Determine the actions for this URL
2196 #ifdef FEATURE_TOGGLE
2197 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2199 /* Most compatible set of actions (i.e. none) */
2200 init_current_action(csp->action);
2203 #endif /* ndef FEATURE_TOGGLE */
2205 get_url_actions(csp, http);
2209 * Save a copy of the original request for logging
2211 http->ocmd = strdup(http->cmd);
2213 if (http->ocmd == NULL)
2215 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2218 enlist(csp->headers, http->cmd);
2220 /* Append the previously read headers */
2221 list_append_list_unique(csp->headers, headers);
2222 destroy_list(headers);
2224 err = sed(csp, FILTER_CLIENT_HEADERS);
2225 if (JB_ERR_OK != err)
2227 assert(err == JB_ERR_PARSE);
2228 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2230 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2232 /* Check request line for rewrites. */
2233 if ((NULL == csp->headers->first->str)
2234 || (strcmp(http->cmd, csp->headers->first->str) &&
2235 (JB_ERR_OK != change_request_destination(csp))))
2238 * A header filter broke the request line - bail out.
2240 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2241 /* XXX: Use correct size */
2242 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2243 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2245 free_http_request(http);
2249 /* decide how to route the HTTP request */
2250 fwd = forward_url(csp, http);
2253 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2254 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2258 * build the http request to send to the server
2259 * we have to do one of the following:
2261 * create = use the original HTTP request to create a new
2262 * HTTP request that has either the path component
2263 * without the http://domainspec (w/path) or the
2264 * full orininal URL (w/url)
2265 * Note that the path and/or the HTTP version may
2266 * have been altered by now.
2268 * connect = Open a socket to the host:port of the server
2269 * and short-circuit server and client socket.
2271 * pass = Pass the request unchanged if forwarding a CONNECT
2272 * request to a parent proxy. Note that we'll be sending
2273 * the CFAIL message ourselves if connecting to the parent
2274 * fails, but we won't send a CSUCCEED message if it works,
2275 * since that would result in a double message (ours and the
2276 * parent's). After sending the request to the parent, we simply
2279 * here's the matrix:
2282 * +--------+--------+
2284 * 0 | create | connect|
2286 * Forwarding +--------+--------+
2288 * 1 | create | pass |
2290 * +--------+--------+
2294 if (http->ssl && connect_port_is_forbidden(csp))
2296 const char *acceptable_connect_ports =
2297 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2298 assert(NULL != acceptable_connect_ports);
2299 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2300 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2301 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2302 csp->action->flags |= ACTION_BLOCK;
2308 freez(csp->headers->first->str);
2309 build_request_line(csp, fwd, &csp->headers->first->str);
2312 hdr = list_to_text(csp->headers);
2315 /* FIXME Should handle error properly */
2316 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2320 * We have a request. Check if one of the crunchers wants it.
2322 if (crunch_response_triggered(csp, crunchers_all))
2325 * Yes. The client got the crunch response
2326 * and we are done here after cleaning up.
2329 list_remove_all(csp->headers);
2335 * The headers can't be removed earlier because
2336 * they were still needed for the referrer check
2337 * in case of CGI crunches.
2339 * XXX: Would it be worth to move the referrer check
2340 * into client_referrer() and set a flag if it's trusted?
2342 list_remove_all(csp->headers);
2344 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2346 if (fwd->forward_host)
2348 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2349 fwd->forward_host, fwd->forward_port, http->hostport);
2353 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2356 /* here we connect to the server, gateway, or the forwarder */
2358 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2359 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2361 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2362 forwarded_connect_retries, http->hostport);
2365 if (csp->sfd == JB_INVALID_SOCKET)
2367 if (fwd->type != SOCKS_NONE)
2370 rsp = error_response(csp, "forwarding-failed", errno);
2372 else if (errno == EINVAL)
2374 rsp = error_response(csp, "no-such-domain", errno);
2378 rsp = error_response(csp, "connect-failed", errno);
2379 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2384 /* Write the answer to the client */
2387 send_crunch_response(csp, rsp);
2394 if (fwd->forward_host || (http->ssl == 0))
2396 /* write the client's (modified) header to the server
2397 * (along with anything else that may be in the buffer)
2400 if (write_socket(csp->sfd, hdr, strlen(hdr))
2401 || (flush_socket(csp->sfd, csp->iob) < 0))
2403 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2406 rsp = error_response(csp, "connect-failed", errno);
2410 send_crunch_response(csp, rsp);
2420 * We're running an SSL tunnel and we're not forwarding,
2421 * so just send the "connect succeeded" message to the
2422 * client, flush the rest, and get out of the way.
2424 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2432 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2434 /* we're finished with the client's header */
2437 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2439 /* pass data between the client and server
2440 * until one or the other shuts down the connection.
2449 * FD_ZERO here seems to point to an errant macro which crashes.
2450 * So do this by hand for now...
2452 memset(&rfds,0x00,sizeof(fd_set));
2456 FD_SET(csp->cfd, &rfds);
2457 FD_SET(csp->sfd, &rfds);
2459 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2463 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2467 /* this is the body of the browser's request
2468 * just read it and write it.
2471 if (FD_ISSET(csp->cfd, &rfds))
2473 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2477 break; /* "game over, man" */
2480 if (write_socket(csp->sfd, buf, (size_t)len))
2482 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2489 * The server wants to talk. It could be the header or the body.
2490 * If `hdr' is null, then it's the header otherwise it's the body.
2491 * FIXME: Does `hdr' really mean `host'? No.
2495 if (FD_ISSET(csp->sfd, &rfds))
2498 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2502 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2504 if (http->ssl && (fwd->forward_host == NULL))
2507 * Just hang up. We already confirmed the client's CONNECT
2508 * request with status code 200 and unencrypted content is
2509 * no longer welcome.
2511 log_error(LOG_LEVEL_ERROR,
2512 "CONNECT already confirmed. Unable to tell the client about the problem.");
2515 else if (byte_count)
2518 * Just hang up. We already transmitted the original headers
2519 * and parts of the original content and therefore missed the
2520 * chance to send an error message (without risking data corruption).
2522 * XXX: we could retry with a fancy range request here.
2524 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2525 "Unable to tell the client about the problem.");
2529 rsp = error_response(csp, "connect-failed", errno);
2533 send_crunch_response(csp, rsp);
2539 /* Add a trailing zero. This lets filter_popups
2540 * use string operations.
2544 /* Normally, this would indicate that we've read
2545 * as much as the server has sent us and we can
2546 * close the client connection. However, Microsoft
2547 * in its wisdom has released IIS/5 with a bug that
2548 * prevents it from sending the trailing \r\n in
2549 * a 302 redirect header (and possibly other headers).
2550 * To work around this if we've haven't parsed
2551 * a full header we'll append a trailing \r\n
2552 * and see if this now generates a valid one.
2554 * This hack shouldn't have any impacts. If we've
2555 * already transmitted the header or if this is a
2556 * SSL connection, then we won't bother with this
2557 * hack. So we only work on partially received
2558 * headers. If we append a \r\n and this still
2559 * doesn't generate a valid header, then we won't
2560 * transmit anything to the client.
2565 if (server_body || http->ssl)
2568 * If we have been buffering up the document,
2569 * now is the time to apply content modification
2570 * and send the result to the client.
2574 p = execute_content_filter(csp, content_filter);
2576 * If the content filter fails, use the original
2577 * buffer and length.
2578 * (see p != NULL ? p : csp->iob->cur below)
2582 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2585 if (JB_ERR_OK != update_server_headers(csp))
2587 log_error(LOG_LEVEL_FATAL,
2588 "Failed to update server headers. after filtering.");
2591 hdr = list_to_text(csp->headers);
2594 /* FIXME Should handle error properly */
2595 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2598 if (write_socket(csp->cfd, hdr, strlen(hdr))
2599 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2601 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2611 break; /* "game over, man" */
2615 * This is NOT the body, so
2616 * Let's pretend the server just sent us a blank line.
2618 snprintf(buf, sizeof(buf), "\r\n");
2619 len = (int)strlen(buf);
2622 * Now, let the normal header parsing algorithm below do its
2623 * job. If it fails, we'll exit instead of continuing.
2630 * If this is an SSL connection or we're in the body
2631 * of the server document, just write it to the client,
2632 * unless we need to buffer the body for later content-filtering
2635 if (server_body || http->ssl)
2640 * If there is no memory left for buffering the content, or the buffer limit
2641 * has been reached, switch to non-filtering mode, i.e. make & write the
2642 * header, flush the iob and buf, and get out of the way.
2644 if (add_to_iob(csp, buf, len))
2649 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2651 hdr = list_to_text(csp->headers);
2655 * Memory is too tight to even generate the header.
2656 * Send our static "Out-of-memory" page.
2658 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2659 rsp = cgi_error_memory();
2660 send_crunch_response(csp, rsp);
2664 hdrlen = strlen(hdr);
2666 if (write_socket(csp->cfd, hdr, hdrlen)
2667 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2668 || (write_socket(csp->cfd, buf, (size_t)len)))
2670 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2677 * Reset the byte_count to the amount of bytes
2678 * we just flushed. len will be added a few lines below,
2679 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2681 byte_count = (size_t)flushed;
2683 content_filter = NULL;
2689 if (write_socket(csp->cfd, buf, (size_t)len))
2691 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2695 byte_count += (size_t)len;
2700 /* we're still looking for the end of the
2701 * server's header ... (does that make header
2702 * parsing an "out of body experience" ?
2706 * buffer up the data we just read. If that fails,
2707 * there's little we can do but send our static
2708 * out-of-memory page.
2710 if (add_to_iob(csp, buf, len))
2712 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2713 rsp = cgi_error_memory();
2714 send_crunch_response(csp, rsp);
2719 /* Convert iob into something sed() can digest */
2720 if (JB_ERR_PARSE == get_server_headers(csp))
2724 /* Well, we tried our MS IIS/5
2725 * hack and it didn't work.
2726 * The header is incomplete
2727 * and there isn't anything
2728 * we can do about it.
2730 log_error(LOG_LEVEL_INFO,
2731 "MS IIS5 hack didn't produce valid headers.");
2736 /* Since we have to wait for
2737 * more from the server before
2738 * we can parse the headers
2739 * we just continue here.
2745 /* Did we actually get anything? */
2746 if (NULL == csp->headers->first)
2748 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2749 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2750 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2751 free_http_request(http);
2755 assert(csp->headers->first->str);
2757 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2758 strncmpic(csp->headers->first->str, "ICY", 3))
2761 * It doesn't look like a HTTP (or Shoutcast) response:
2762 * tell the client and log the problem.
2764 if (strlen(csp->headers->first->str) > 30)
2766 csp->headers->first->str[30] = '\0';
2768 log_error(LOG_LEVEL_ERROR,
2769 "Invalid server or forwarder response. Starts with: %s",
2770 csp->headers->first->str);
2771 log_error(LOG_LEVEL_CLF,
2772 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2773 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2774 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2775 free_http_request(http);
2780 * We have now received the entire server header,
2781 * filter it and send the result to the client
2783 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2785 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2787 hdr = list_to_text(csp->headers);
2790 /* FIXME Should handle error properly */
2791 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2794 if (crunch_response_triggered(csp, crunchers_light))
2797 * One of the tags created by a server-header
2798 * tagger triggered a crunch. We already
2799 * delivered the crunch response to the client
2800 * and are done here after cleaning up.
2805 /* Buffer and pcrs filter this if appropriate. */
2807 if (!http->ssl) /* We talk plaintext */
2809 content_filter = get_filter_function(csp);
2812 * Only write if we're not buffering for content modification
2814 if (!content_filter)
2816 /* write the server's (modified) header to
2817 * the client (along with anything else that
2818 * may be in the buffer)
2821 if (write_socket(csp->cfd, hdr, strlen(hdr))
2822 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2824 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2826 /* the write failed, so don't bother
2827 * mentioning it to the client...
2828 * it probably can't hear us anyway.
2834 byte_count += (size_t)len;
2837 /* we're finished with the server's header */
2842 /* If this was a MS IIS/5 hack then it means
2843 * the server has already closed the
2844 * connection. Nothing more to read. Time
2849 log_error(LOG_LEVEL_INFO,
2850 "Closed server connection detected with MS IIS5 hack enabled.");
2857 return; /* huh? we should never get here */
2860 if (csp->content_length == 0)
2863 * If Privoxy didn't recalculate the
2864 * Content-Lenght, byte_count is still
2867 csp->content_length = byte_count;
2870 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2871 csp->ip_addr_str, http->ocmd, csp->content_length);
2875 /*********************************************************************
2879 * Description : This is little more than chat. We only "serve" to
2880 * to close any socket that chat may have opened.
2883 * 1 : csp = Current client state (buffers, headers, etc...)
2887 *********************************************************************/
2889 void serve(struct client_state *csp)
2890 #else /* ifndef AMIGA */
2891 static void serve(struct client_state *csp)
2892 #endif /* def AMIGA */
2895 close_socket(csp->cfd);
2897 if (csp->sfd != JB_INVALID_SOCKET)
2899 close_socket(csp->sfd);
2902 csp->flags &= ~CSP_FLAG_ACTIVE;
2908 /*********************************************************************
2910 * Function : server_thread
2912 * Description : We only exist to call `serve' in a threaded environment.
2915 * 1 : data = Current client state (buffers, headers, etc...)
2917 * Returns : Always 0.
2919 *********************************************************************/
2920 static int32 server_thread(void *data)
2922 serve((struct client_state *) data);
2929 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2930 /*********************************************************************
2934 * Description : Print usage info & exit.
2936 * Parameters : Pointer to argv[0] for identifying ourselves
2940 *********************************************************************/
2941 static void usage(const char *myname)
2943 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2947 #endif /* defined(unix) */
2950 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2951 #endif /* defined(unix) */
2952 "[--version] [configfile]\n"
2953 "Aborting\n", myname);
2958 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2961 #ifdef FEATURE_PTHREAD
2962 /*********************************************************************
2964 * Function : privoxy_mutex_lock
2966 * Description : Locks a mutex using pthread_mutex_lock.
2969 * 1 : mutex = The mutex to lock.
2971 * Returns : Void, exits in case of errors.
2973 *********************************************************************/
2974 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
2976 int err = pthread_mutex_lock(mutex);
2979 if (mutex != &log_mutex)
2981 log_error(LOG_LEVEL_FATAL,
2982 "Mutex locking failed: %s.\n", strerror(err));
2989 /*********************************************************************
2991 * Function : privoxy_mutex_unlock
2993 * Description : Unlocks a mutex using pthread_mutex_unlock.
2996 * 1 : mutex = The mutex to unlock.
2998 * Returns : Void, exits in case of errors.
3000 *********************************************************************/
3001 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3003 int err = pthread_mutex_unlock(mutex);
3006 if (mutex != &log_mutex)
3008 log_error(LOG_LEVEL_FATAL,
3009 "Mutex unlocking failed: %s.\n", strerror(err));
3016 /*********************************************************************
3018 * Function : privoxy_mutex_init
3020 * Description : Prepares a mutex using pthread_mutex_init.
3023 * 1 : mutex = The mutex to initialize.
3025 * Returns : Void, exits in case of errors.
3027 *********************************************************************/
3028 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3030 int err = pthread_mutex_init(mutex, 0);
3033 printf("Fatal error. Mutex initialization failed: %s.\n",
3038 #endif /* def FEATURE_PTHREAD */
3041 /*********************************************************************
3043 * Function : initialize_mutexes
3045 * Description : Prepares mutexes if mutex support is available.
3049 * Returns : Void, exits in case of errors.
3051 *********************************************************************/
3052 static void initialize_mutexes(void)
3054 #ifdef FEATURE_PTHREAD
3056 * Prepare global mutex semaphores
3058 privoxy_mutex_init(&log_mutex);
3060 privoxy_mutex_init(&log_init_mutex);
3063 * XXX: The assumptions below are a bit naive
3064 * and can cause locks that aren't necessary.
3066 * For example older FreeBSD versions (< 6.x?)
3067 * have no gethostbyname_r, but gethostbyname is
3070 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3071 privoxy_mutex_init(&resolver_mutex);
3072 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3074 * XXX: should we use a single mutex for
3075 * localtime() and gmtime() as well?
3077 #ifndef HAVE_GMTIME_R
3078 privoxy_mutex_init(&gmtime_mutex);
3079 #endif /* ndef HAVE_GMTIME_R */
3081 #ifndef HAVE_LOCALTIME_R
3082 privoxy_mutex_init(&localtime_mutex);
3083 #endif /* ndef HAVE_GMTIME_R */
3086 privoxy_mutex_init(&rand_mutex);
3087 #endif /* ndef HAVE_RANDOM */
3088 #endif /* FEATURE_PTHREAD */
3091 * TODO: mutex support for mingw32 would be swell.
3097 /*********************************************************************
3101 * Description : Load the config file and start the listen loop.
3102 * This function is a lot more *sane* with the `load_config'
3103 * and `listen_loop' functions; although it stills does
3104 * a *little* too much for my taste.
3107 * 1 : argc = Number of parameters (including $0).
3108 * 2 : argv = Array of (char *)'s to the parameters.
3110 * Returns : 1 if : can't open config file, unrecognized directive,
3111 * stats requested in multi-thread mode, can't open the
3112 * log file, can't open the jar file, listen port is invalid,
3113 * any load fails, and can't bind port.
3115 * Else main never returns, the process must be signaled
3116 * to terminate execution. Or, on Windows, use the
3117 * "File", "Exit" menu option.
3119 *********************************************************************/
3121 int real_main(int argc, const char *argv[])
3123 int main(int argc, const char *argv[])
3127 unsigned int random_seed;
3129 struct passwd *pw = NULL;
3130 struct group *grp = NULL;
3133 char *pre_chroot_nslookup_to_load_resolver = NULL;
3140 #if !defined(_WIN32)
3148 * Parse the command line arguments
3150 * XXX: simply printing usage information in case of
3151 * invalid arguments isn't particularly user friendly.
3153 while (++argc_pos < argc)
3156 /* Check to see if the service must be installed or uninstalled */
3157 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3159 const char *pName = argv[argc_pos] + 9;
3162 exit( (install_service(pName)) ? 0 : 1 );
3164 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3166 const char *pName = argv[argc_pos] + 11;
3169 exit((uninstall_service(pName)) ? 0 : 1);
3171 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3173 bRunAsService = TRUE;
3174 w32_set_service_cwd();
3175 atexit(w32_service_exit_notify);
3178 #endif /* defined(_WIN32) */
3181 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3183 if (strcmp(argv[argc_pos], "--help") == 0)
3188 else if(strcmp(argv[argc_pos], "--version") == 0)
3190 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3196 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3201 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3203 if (++argc_pos == argc) usage(argv[0]);
3204 pidfile = strdup(argv[argc_pos]);
3207 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3209 if (++argc_pos == argc) usage(argv[argc_pos]);
3211 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3214 if (NULL == (grp = getgrnam(p)))
3216 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3220 if (NULL == (pw = getpwnam(argv[argc_pos])))
3222 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3225 if (p != NULL) *--p = '\0';
3228 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3230 if (++argc_pos == argc) usage(argv[0]);
3231 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3234 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3238 #endif /* defined(unix) */
3240 else if (argc_pos + 1 != argc)
3243 * This is neither the last command line
3244 * option, nor was it recognized before,
3245 * therefore it must be invalid.
3251 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3253 configfile = argv[argc_pos];
3256 } /* -END- while (more arguments) */
3259 if ( *configfile != '/' )
3261 char cwd[BUFFER_SIZE];
3263 size_t abs_file_size;
3265 /* make config-filename absolute here */
3266 if (NULL == getcwd(cwd, sizeof(cwd)))
3268 perror("failed to get current working directory");
3273 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3274 basedir = strdup(cwd);
3276 if (NULL == basedir ||
3277 NULL == (abs_file = malloc(abs_file_size)))
3279 perror("malloc failed");
3282 strlcpy(abs_file, basedir, abs_file_size);
3283 strlcat(abs_file, "/", abs_file_size );
3284 strlcat(abs_file, configfile, abs_file_size);
3285 configfile = abs_file;
3287 #endif /* defined unix */
3291 clients->next = NULL;
3293 /* XXX: factor out initialising after the next stable release. */
3296 #elif defined(_WIN32)
3300 /* Prepare mutexes if supported and necessary. */
3301 initialize_mutexes();
3303 /* Enable logging until further notice. */
3304 init_log_module(Argv[0]);
3306 random_seed = (unsigned int)time(NULL);
3308 srandom(random_seed);
3309 #elif defined (_WIN32)
3311 * See pick_from_range() in miscutil.c for details.
3313 log_error(LOG_LEVEL_INFO,
3314 "No thread-safe PRNG implemented for your platform. "
3315 "Using weak \'randomization\' factor which will "
3316 "limit the already questionable usefulness of "
3317 "header-time-randomizing actions (disabled by default).");
3320 #endif /* ifdef HAVE_RANDOM */
3323 * Unix signal handling
3325 * Catch the abort, interrupt and terminate signals for a graceful exit
3326 * Catch the hangup signal so the errlog can be reopened.
3327 * Ignore the broken pipe signals (FIXME: Why?)
3329 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3332 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3333 const int ignored_signals[] = { SIGPIPE, 0 };
3335 for (idx = 0; catched_signals[idx] != 0; idx++)
3337 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3338 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3340 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3341 #endif /* ifdef sun */
3343 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3347 for (idx = 0; ignored_signals[idx] != 0; idx++)
3349 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3351 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3356 #else /* ifdef _WIN32 */
3357 # ifdef _WIN_CONSOLE
3359 * We *are* in a windows console app.
3360 * Print a verbose messages about FAQ's and such
3362 printf("%s", win32_blurb);
3363 # endif /* def _WIN_CONSOLE */
3364 #endif /* def _WIN32 */
3367 /* Initialize the CGI subsystem */
3368 cgi_init_error_messages();
3371 * If runnig on unix and without the --nodaemon
3372 * option, become a daemon. I.e. fork, detach
3373 * from tty and get process group leadership
3386 if ( pid < 0 ) /* error */
3391 else if ( pid != 0 ) /* parent */
3396 * must check for errors
3397 * child died due to missing files aso
3400 wpid = waitpid( pid, &status, WNOHANG );
3409 /* Should be more portable, but not as well tested */
3414 #else /* ndef __FreeBSD__ */
3416 #endif /* ndef __FreeBSD__ */
3417 fd = open("/dev/tty", O_RDONLY);
3420 /* no error check here */
3421 ioctl( fd, TIOCNOTTY,0 );
3426 * stderr (fd 2) will be closed later on, when the
3427 * log file has been parsed.
3434 } /* -END- if (!no_daemon) */
3437 * As soon as we have written the PID file, we can switch
3438 * to the user and group ID indicated by the --user option
3444 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3446 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3452 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3454 /* Read the time zone file from /etc before doing chroot. */
3456 if (NULL != pre_chroot_nslookup_to_load_resolver
3457 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3459 /* Initialize resolver library. */
3460 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3462 if (chroot(pw->pw_dir) < 0)
3464 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3468 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3471 if (setuid(pw->pw_uid))
3473 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3477 char putenv_dummy[64];
3479 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3480 if (putenv(putenv_dummy) != 0)
3482 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3485 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3486 if (putenv(putenv_dummy) != 0)
3488 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3494 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3497 #endif /* defined unix */
3500 /* This will be FALSE unless the command line specified --service
3504 /* Yup, so now we must attempt to establish a connection
3505 * with the service dispatcher. This will only work if this
3506 * process was launched by the service control manager to
3507 * actually run as a service. If this isn't the case, i've
3508 * known it take around 30 seconds or so for the call to return.
3511 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3512 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3514 /* Service has run, and at this point is now being stopped, so just return */
3519 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3521 /* An error occurred. Usually it's because --service was wrongly specified
3522 * and we were unable to connect to the Service Control Dispatcher because
3523 * it wasn't expecting us and is therefore not listening.
3525 * For now, just continue below to call the listen_loop function.
3528 #endif /* def _WIN32 */
3538 /*********************************************************************
3540 * Function : bind_port_helper
3542 * Description : Bind the listen port. Handles logging, and aborts
3546 * 1 : config = Privoxy configuration. Specifies port
3549 * Returns : Port that was opened.
3551 *********************************************************************/
3552 static jb_socket bind_port_helper(struct configuration_spec * config)
3557 if (config->haddr == NULL)
3559 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3564 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3565 config->hport, config->haddr);
3568 result = bind_port(config->haddr, config->hport, &bfd);
3575 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3576 "There may be another Privoxy or some other "
3577 "proxy running on port %d",
3578 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3579 config->hport, config->hport);
3582 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3583 "The hostname is not resolvable",
3584 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3587 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3588 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3591 /* shouldn't get here */
3592 return JB_INVALID_SOCKET;
3595 config->need_bind = 0;
3602 /* Without this simple workaround we get this compiler warning from _beginthread
3603 * warning C4028: formal parameter 1 different from declaration
3605 void w32_service_listen_loop(void *p)
3609 #endif /* def _WIN32 */
3612 /*********************************************************************
3614 * Function : listen_loop
3616 * Description : bind the listen port and enter a "FOREVER" listening loop.
3622 *********************************************************************/
3623 static void listen_loop(void)
3625 struct client_state *csp = NULL;
3627 struct configuration_spec * config;
3629 config = load_config();
3631 bfd = bind_port_helper(config);
3633 #ifdef FEATURE_GRACEFUL_TERMINATION
3634 while (!g_terminate)
3639 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3640 while (waitpid(-1, NULL, WNOHANG) > 0)
3642 /* zombie children */
3644 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3647 * Free data that was used by died threads
3653 * Re-open the errlog after HUP signal
3655 if (received_hup_signal)
3657 init_error_log(Argv[0], config->logfile);
3658 received_hup_signal = 0;
3662 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3664 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3668 csp->flags |= CSP_FLAG_ACTIVE;
3669 csp->sfd = JB_INVALID_SOCKET;
3671 csp->config = config = load_config();
3673 if ( config->need_bind )
3676 * Since we were listening to the "old port", we will not see
3677 * a "listen" param change until the next IJB request. So, at
3678 * least 1 more request must be made for us to find the new
3679 * setting. I am simply closing the old socket and binding the
3682 * Which-ever is correct, we will serve 1 more page via the
3683 * old settings. This should probably be a "show-proxy-args"
3684 * request. This should not be a so common of an operation
3685 * that this will hurt people's feelings.
3690 bfd = bind_port_helper(config);
3693 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3695 if (!accept_connection(csp, bfd))
3697 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3710 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3713 #ifdef FEATURE_TOGGLE
3714 if (global_toggle_state)
3715 #endif /* def FEATURE_TOGGLE */
3717 csp->flags |= CSP_FLAG_TOGGLED_ON;
3720 if (run_loader(csp))
3722 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3723 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3727 if (block_acl(NULL,csp))
3729 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3730 close_socket(csp->cfd);
3734 #endif /* def FEATURE_ACL */
3736 /* add it to the list of clients */
3737 csp->next = clients->next;
3738 clients->next = csp;
3740 if (config->multi_threaded)
3744 /* this is a switch () statment in the C preprocessor - ugh */
3745 #undef SELECTED_ONE_OPTION
3747 /* Use Pthreads in preference to native code */
3748 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3749 #define SELECTED_ONE_OPTION
3751 pthread_t the_thread;
3752 pthread_attr_t attrs;
3754 pthread_attr_init(&attrs);
3755 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3756 errno = pthread_create(&the_thread, &attrs,
3757 (void * (*)(void *))serve, csp);
3758 child_id = errno ? -1 : 0;
3759 pthread_attr_destroy(&attrs);
3763 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3764 #define SELECTED_ONE_OPTION
3765 child_id = _beginthread(
3766 (void (*)(void *))serve,
3771 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3772 #define SELECTED_ONE_OPTION
3773 child_id = _beginthread(
3774 (void(* _Optlink)(void*))serve,
3780 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3781 #define SELECTED_ONE_OPTION
3783 thread_id tid = spawn_thread
3784 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3786 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3788 child_id = (int) tid;
3797 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3798 #define SELECTED_ONE_OPTION
3799 csp->cfd = ReleaseSocket(csp->cfd, -1);
3802 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3803 NP_Output, Output(),
3804 NP_CloseOutput, FALSE,
3805 NP_Name, (ULONG)"privoxy child",
3809 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3810 NP_Output, Output(),
3811 NP_CloseOutput, FALSE,
3812 NP_Name, (ULONG)"privoxy child",
3813 NP_StackSize, 200*1024,
3819 ((struct Task *)child_id)->tc_UserData = csp;
3820 Signal((struct Task *)child_id, SIGF_SINGLE);
3825 #if !defined(SELECTED_ONE_OPTION)
3828 /* This block is only needed when using fork().
3829 * When using threads, the server thread was
3830 * created and run by the call to _beginthread().
3832 if (child_id == 0) /* child */
3835 #ifdef FEATURE_TOGGLE
3836 int inherited_toggle_state = global_toggle_state;
3837 #endif /* def FEATURE_TOGGLE */
3842 * If we've been toggled or we've blocked the request, tell Mom
3845 #ifdef FEATURE_TOGGLE
3846 if (inherited_toggle_state != global_toggle_state)
3848 rc |= RC_FLAG_TOGGLED;
3850 #endif /* def FEATURE_TOGGLE */
3852 #ifdef FEATURE_STATISTICS
3853 if (csp->flags & CSP_FLAG_REJECTED)
3855 rc |= RC_FLAG_BLOCKED;
3857 #endif /* ndef FEATURE_STATISTICS */
3861 else if (child_id > 0) /* parent */
3863 /* in a fork()'d environment, the parent's
3864 * copy of the client socket and the CSP
3868 #if !defined(_WIN32) && !defined(__CYGWIN__)
3870 wait( &child_status );
3873 * Evaluate child's return code: If the child has
3874 * - been toggled, toggle ourselves
3875 * - blocked its request, bump up the stats counter
3878 #ifdef FEATURE_TOGGLE
3879 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3881 global_toggle_state = !global_toggle_state;
3883 #endif /* def FEATURE_TOGGLE */
3885 #ifdef FEATURE_STATISTICS
3887 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3891 #endif /* def FEATURE_STATISTICS */
3893 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3894 close_socket(csp->cfd);
3895 csp->flags &= ~CSP_FLAG_ACTIVE;
3899 #undef SELECTED_ONE_OPTION
3900 /* end of cpp switch () */
3902 if (child_id < 0) /* failed */
3904 char buf[BUFFER_SIZE];
3906 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3908 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3910 write_socket(csp->cfd, buf, strlen(buf));
3911 close_socket(csp->cfd);
3912 csp->flags &= ~CSP_FLAG_ACTIVE;
3923 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3925 /* Clean up. Aim: free all memory (no leaks) */
3926 #ifdef FEATURE_GRACEFUL_TERMINATION
3928 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3930 unload_current_config_file();
3931 unload_current_actions_file();
3932 unload_current_re_filterfile();
3933 #ifdef FEATURE_TRUST
3934 unload_current_trust_file();
3937 if (config->multi_threaded)
3944 } while ((clients->next != NULL) && (--i > 0));
3948 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3959 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3960 /* Cleanup - remove taskbar icon etc. */
3965 #endif /* FEATURE_GRACEFUL_TERMINATION */