1 const char jcc_rcs[] = "$Id: jcc.c,v 1.182 2008/06/27 11:13:56 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 1.182 2008/06/27 11:13:56 fabiankeil
37 * Fix possible NULL-pointer dereference reported
38 * by din_a4 in #2003937. Pointy hat to me.
40 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
41 * Streamline sed()'s prototype and declare
42 * the header parse and add structures static.
44 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
45 * - Mark csp as immutable for send_crunch_response().
46 * - Fix comment spelling.
48 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
49 * Factor update_server_headers() out of sed(), ditch the
50 * first_run hack and make server_patterns_light static.
52 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
53 * Don't provide get_header() with the whole client state
54 * structure when it only needs access to csp->iob.
56 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
57 * Make the "read the rest of the headers" loop a bit more readable.
59 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
60 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
61 * - Remove useless comment.
63 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
64 * Fix comment grammar.
66 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
67 * Remove the pointless buffer in client_protocol_is_unsupported().
69 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
70 * Least-effort fix for bug #1821930 (reported by Lee):
71 * If the response doesn't look like HTTP,
72 * tell the client and log the problem.
74 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
75 * Don't pass the whole csp structure to flush_socket()
76 * when it only needs a file descriptor and a buffer.
78 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
79 * Remove kill-popups action.
81 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
82 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
84 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
85 * Remove the treat-forbidden-connects-like-blocks action. We now
86 * use the "blocked" page for forbidden CONNECT requests by default.
88 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
89 * Also use shiny new connect_port_is_forbidden() in jcc.c.
91 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
92 * Rename url_actions() to get_url_actions() and let it
93 * use the standard parameter ordering.
95 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
96 * Let forward_url() use the standard parameter ordering
97 * and mark its second parameter immutable.
99 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
100 * Remove the "Listening ... for local connections only" log message.
101 * Whether or not remote connections are able to reach Privoxy is up
102 * to the operating system.
104 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
105 * Prevent the log messages for CONNECT requests to unacceptable
106 * ports from printing the limit-connect argument as [null] if
107 * limit-connect hasn't been explicitly enabled.
109 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
110 * Make sure all console-mode apps get a usage() instance
112 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
113 * Reword NO_SERVER_DATA_RESPONSE to make it harder
114 * to misunderstand what the message is all about.
116 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
117 * Unbreak trustfile which previously didn't work without
118 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
120 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
121 * Plug memory leak. Spotted by Valgrind, triggered by
122 * Privoxy-Regression-Test feeding proxyfuzz.py.
124 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
125 * In the HTTP snipplets, refer to the client as client.
127 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
128 * Emit a log message when activating the MS IIS5 hack.
130 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
131 * Log the "weak randomization factor" warning only
132 * once for mingw32 and provide some more details.
134 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
135 * Initialize log module after initializing mutexes, future
136 * deadlocks in that code should now work cross-platform.
138 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
139 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
140 * Reported by Bert van Leeuwen in BR#1818808.
142 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
143 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
145 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
146 * When in daemon mode, close stderr after the configuration file has been
147 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
149 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
150 * - Fix a crash when parsing invalid requests whose first header
151 * is rejected by get_header(). Regression (re?)introduced
152 * in r1.143 by yours truly.
153 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
154 * client_cookie_adder() to make sure send-vanilla-wafer can be
155 * controlled through tags (and thus regression-tested).
157 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
158 * - Move get_filter_function() from jcc.c to filters.c
159 * so the filter functions can be static.
160 * - Don't bother filtering body-less responses.
162 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
163 * Execute content filters through execute_content_filter().
165 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
166 * Initialize req to NULL to make sure it's defined if the
167 * first read_socket() call fails. Reported by icmp30.
169 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
170 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
171 * extensive comments moved to user manual.
173 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
174 * Don't crash if a broken header filter wiped out the request line.
176 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
177 * Fix byte_count calculation in case of flushes
178 * and don't parse the server headers a second time.
180 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
181 * - If there's a connection problem after we already forwarded
182 * parts of the original content, just hang up. Fixes BR#1776724.
183 * - Fix warnings about unused code on mingw32.
184 * - In case of flushes, calculate the byte count
185 * less incorrectly (I think).
187 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
188 * Add some more prototypes for static functions.
190 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
191 * Comment out request_contains_null_bytes() until it's used again.
193 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
194 * #1763173 from Stefan Huehner: s@const static@static const@
195 * and declare some more functions static.
197 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
198 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
199 * and explain why they were rejected in the first place.
200 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
201 * CONNECT requests. The request line was missing.
202 * - Add two more XXX reminders as we don't have enough already.
204 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
205 * As Hal noticed, checking dispatch_cgi() as the last cruncher
206 * looks like a bug if CGI requests are blocked unintentionally,
207 * so don't do it unless the user enabled the new config option
208 * "allow-cgi-request-crunching".
210 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
211 * - Allow to rewrite the request destination behind the client's back.
212 * - Turn the weird-looking unconditional for loop that
213 * reads the client request into a conditional while loop.
214 * Move the stuff that only runs once out of the loop.
215 * - Move parts of chat(), server_content_type() and the
216 * necessary stuff to fix BR#1750917 into get_filter_function().
218 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
219 * Temporary workaround for BR#1730105.
221 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
222 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
223 * deadlocks and crashes on OpenBSD and possibly other OS with neither
224 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
225 * Thanks to Ralf Horstmann for report and solution.
227 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
228 * Add forward-override{} to change the forwarding settings through
229 * action sections. This is mainly interesting to forward different
230 * clients differently (for example based on User-Agent or request
233 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
234 * - Let usage() mention the --chroot parameter.
235 * - Use read_socket() consistently and always leave
236 * the last buffer byte alone, even in cases where
237 * null termination (currently) doesn't matter.
239 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
240 * - Fix config file loading on Unix if no config file is specified.
241 * Since r1.97 Privoxy would always interpret the last argument as
242 * config file, even if it's a valid command line option.
243 * - Abort in case of unrecognized command line options. Closes #1719696.
244 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
245 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
247 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
248 * - Don't rerun crunchers that only depend on the request URL.
249 * - Don't count redirects and CGI requests as "blocked requests".
251 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
252 * Support crunching based on tags created by server-header taggers.
254 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
255 * Make HTTP snippets static (again). Add a Content-Type for those
256 * with content so the browser doesn't guess it based on the URL.
258 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
259 * Move crunching and request line rebuilding out of chat().
261 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
262 * Introduce tags as alternative way to specify which
263 * actions apply to a request. At the moment tags can be
264 * created based on client and server headers.
266 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
267 * Don't CLF-log CONNECT requests twice.
269 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
270 * Log the source address for ACL-related connection drops.
272 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
273 * New config option: enforce-blocks.
275 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
276 * - Move null byte check into separate function.
277 * - Don't confuse the client with error pages
278 * if a CONNECT request was already confirmed.
280 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
281 * Speed up NULL byte escaping and only log the complete
282 * NULL byte requests with header debugging enabled.
284 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
285 * Answer requests that contain NULL bytes with
286 * a custom response instead of waiting for more
287 * data until the client eventually hangs up.
289 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
290 * - Move delivery and logging of crunched responses
291 * from chat() into send_crunch_response().
292 * - Display the reason for generating http_responses.
293 * - Log the content length for LOG_LEVEL_CLF correctly
294 * (still incorrect for some fixed responses).
295 * - Reword an incorrect comment about
296 * treat-forbidden-connects-like-blocks violating
298 * - Add some log messages.
300 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
301 * Move mutex initialization into separate
302 * function and exit in case of errors.
304 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
305 * - Start to reduce chat()'s line count and move
306 * parts of it into separate functions.
307 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
309 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
310 * - Add Proxy-Agent header to HTTP snippets that are
311 * supposed to reach HTTP clients only.
312 * - Made a few CONNECT log messages more descriptive.
313 * - Catch completely empty server responses (as seen
314 * with Tor's fake ".noconnect" top level domain).
315 * - Use shiny new "forwarding-failed" template for socks errors.
317 * Revision 1.118 2007/01/07 07:43:43 joergs
318 * AmigaOS4 support added.
320 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
321 * Added config option accept-intercepted-requests
322 * and disabled it by default.
324 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
325 * Reverted parts of my last commit
326 * to keep error handling working.
328 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
329 * Fixed gcc43 conversion warnings.
331 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
332 * Fix -pedantic ISO C warning about converting
333 * from function pointer to object pointer.
335 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
336 * Silence compiler warning I introduced with my last commit.
338 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
339 * Mutex protect rand() if POSIX threading
340 * is used, warn the user if that's not possible
341 * and stop using it on _WIN32 where it could
344 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
345 * Don't prevent core dumps by catching SIGABRT.
346 * It's rude and makes debugging unreasonable painful.
348 * Revision 1.110 2006/12/13 14:52:53 etresoft
349 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
351 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
352 * Privoxy is now able to run as intercepting
353 * proxy in combination with any packet filter
354 * that does the port redirection. The destination
355 * is extracted from the "Host:" header which
356 * should be available for nearly all requests.
358 * Moved HTTP snipplets into jcc.c.
359 * Added error message for gopher proxy requests.
361 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
362 * Only unlink the pidfile if it's actually used.
364 * Change order of interception checks to make
365 * it possible to block or redirect requests for
368 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
369 * Make pthread mutex locking more generic. Instead of
370 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
371 * and use mutex locking unless there is an _r function
372 * available. Better safe than sorry.
374 * Fixes "./configure --disable-pthread" and should result
375 * in less threading-related problems on pthread-using platforms,
376 * but it still doesn't fix BR#1122404.
378 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
379 * Move pthread.h inclusion from jcc.c to jcc.h.
380 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
382 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
383 * Don't exit after receiving the second SIGHUP on Solaris.
385 * Fixes BR 1052235, but the same problem may exist on other
386 * systems. Once 3.0.6 is out we should use sigset()
387 * where available and see if it breaks anything.
389 * Revision 1.104 2006/09/23 13:26:38 roro
390 * Replace TABs by spaces in source code.
392 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
393 * Fix +redirect{}. Didn't work with -fast-redirects.
395 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
396 * Respond with 400 and a short text message
397 * if the client tries to use Privoxy as FTP proxy.
399 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
400 * Make number of retries in case of forwarded-connect problems
401 * a config file option (forwarded-connect-retries) and use 0 as
404 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
405 * Set random(3) seed.
407 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
408 * Follow the OpenBSD port's lead and protect the resolve
409 * functions on OpenBSD as well.
411 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
412 * --user fix. Only use the user as group if no group is specified.
413 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
415 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
416 * Windows service (re-)integration
420 * --install[:service_name]
421 * --uninstall[:service_name]
424 * They work as follows:
425 * --install will create a service for you and then terminate.
426 * By default the service name will be "privoxy" (without the quotes).
427 * However you can run multiple services if you wish, just by adding
428 * a colon and then a name (no spaces).
430 * --uninstall follows the exact same rules a --install.
432 * --service is used when the program is executed by the service
433 * control manager, and in normal circumstances would never be
434 * used as a command line argument.
436 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
437 * Windows service integration
439 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
440 * Incorporate Fabian Keil's patch work:
441 http://www.fabiankeil.de/sourcecode/privoxy/
443 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
444 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
445 * with what was really the latest development (the v_3_0_branch branch)
447 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
448 * Thanks to Jindrich Makovicka for a race condition fix for the log
449 * file. The race condition remains for non-pthread implementations.
450 * Reference patch #1175720.
452 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
453 * Add the ability to check jpeg images for invalid
454 * lengths of comment blocks. Defensive strategy
455 * against the exploit:
456 * Microsoft Security Bulletin MS04-028
457 * Buffer Overrun in JPEG Processing (GDI+) Could
458 * Allow Code Execution (833987)
459 * Enabled with +inspect-jpegs in actions files.
461 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
462 * - Fixed usage info for non-unix platforms
463 * - Fixed small cmdline parsing bug
465 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
466 * Diagnostics: Now preserve the returncode of pthread_create
467 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
469 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
470 * No longer ignore SIGCHLD. Fixes bug #769381
472 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
473 * Close jarfile on graceful exit, remove stray line
475 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
476 * Cosmetics: Killed a warning, a typo and an allocation left at exit
478 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
479 * No longer rely on non-POSIX.1 extensions of getcwd().
482 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
483 * Replaced setenv() by posix-compliant putenv()
484 * Thanks to Neil McCalden (nmcc AT users.sf.net).
486 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
487 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
489 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
490 * Clean-up and extension of improvements for forked mode:
491 * - Child's return code now consists of flags RC_FLAG_*
492 * - Reporting toggle to parent now properly #ifdef'ed
493 * - Children now report blocking to parent. This enables
494 * statistics in forked mode
496 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
497 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
498 * Children now report having been toggled through _exit(17), parents
499 * watch for that code and toggle themselves if found.
501 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
502 * Wrapping all *_r functions (the non-_r versions of them) with
503 * mutex semaphores for OSX. Hopefully this will take care of all
504 * of those pesky crash reports.
506 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
507 * Fixed two mostly harmless mem leaks
509 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
510 * Fix: Head of global clients list now initialized to NULL
512 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
513 * Added basic support for OPTIONS and TRACE HTTP methods:
514 * - New interceptor direct_response() added in chat().
515 * - sed() moved to earlier in the process, so that the
516 * Host: header is evaluated before actions and forwarding
519 * Revision 1.92 2002/05/08 16:00:46 oes
520 * Chat's buffer handling:
521 * - Fixed bug with unchecked out-of-mem conditions
522 * while reading client request & server headers
523 * - No longer predict if the buffer limit will be exceeded
524 * in the next read -- check add_to_iob's new
525 * return code. If buffer couldn't be extended
526 * (policy or out-of-mem) while
527 * - reading from client: abort
528 * - reading server headers: send error page
529 * - buffering server body for filter: flush,
530 * and if that fails: send error page
532 * Revision 1.91 2002/04/08 20:35:58 swa
535 * Revision 1.90 2002/04/02 14:57:28 oes
536 * Made sending wafers independent of FEATURE_COOKIE_JAR
538 * Revision 1.89 2002/03/31 17:18:59 jongfoster
539 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
541 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
542 * More compiler warning message maintenance
544 * Revision 1.87 2002/03/26 22:29:54 swa
545 * we have a new homepage!
547 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
548 * Workaround for closing the jarfile before load_config() comes around again
550 * Revision 1.85 2002/03/24 15:23:33 jongfoster
553 * Revision 1.84 2002/03/24 13:25:43 swa
554 * name change related issues
556 * Revision 1.83 2002/03/16 23:54:06 jongfoster
557 * Adding graceful termination feature, to help look for memory leaks.
558 * If you enable this (which, by design, has to be done by hand
559 * editing config.h) and then go to http://i.j.b/die, then the program
560 * will exit cleanly after the *next* request. It should free all the
561 * memory that was used.
563 * Revision 1.82 2002/03/13 00:27:05 jongfoster
566 * Revision 1.81 2002/03/12 01:42:50 oes
567 * Introduced modular filters
569 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
570 * OS/2 port maintenance:
571 * - Fixed EMX build - it had decayed a little
572 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
573 * substituted a memset for now.
575 * Revision 1.79 2002/03/09 20:03:52 jongfoster
576 * - Making various functions return int rather than size_t.
577 * (Undoing a recent change). Since size_t is unsigned on
578 * Windows, functions like read_socket that return -1 on
579 * error cannot return a size_t.
581 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
582 * crashes, and also frequently caused JB to jump to 100%
583 * CPU and stay there. (Because it thought it had just
584 * read ((unsigned)-1) == 4Gb of data...)
586 * - The signature of write_socket has changed, it now simply
587 * returns success=0/failure=nonzero.
589 * - Trying to get rid of a few warnings --with-debug on
590 * Windows, I've introduced a new type "jb_socket". This is
591 * used for the socket file descriptors. On Windows, this
592 * is SOCKET (a typedef for unsigned). Everywhere else, it's
593 * an int. The error value can't be -1 any more, so it's
594 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
595 * Windows it maps to the #define INVALID_SOCKET.)
597 * - The signature of bind_port has changed.
599 * Revision 1.78 2002/03/08 21:35:04 oes
600 * Added optional group supplement to --user option. Will now use default group of user if no group given
602 * Revision 1.77 2002/03/07 03:52:06 oes
603 * - Fixed compiler warnings etc
604 * - Improved handling of failed DNS lookups
606 * Revision 1.76 2002/03/06 22:54:35 jongfoster
607 * Automated function-comment nitpicking.
609 * Revision 1.75 2002/03/06 10:02:19 oes
610 * Fixed stupid bug when --user was not given
612 * Revision 1.74 2002/03/06 00:49:31 jongfoster
613 * Fixing warning on Windows
614 * Making #ifdefs that refer to the same variable consistently
615 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
617 * Revision 1.73 2002/03/05 23:57:30 hal9
618 * Stray character 's' on line 1618 was breaking build.
620 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
621 * - Re-enable OS/2 building after new parms were added
622 * - Fix false out of memory report when resolving CGI templates when no IP
623 * address is available of failed attempt (a la no such domain)
625 * Revision 1.71 2002/03/05 18:13:56 oes
626 * Added --user option
628 * Revision 1.70 2002/03/05 04:52:42 oes
629 * Deleted non-errlog debugging code
631 * Revision 1.69 2002/03/04 23:50:00 jongfoster
632 * Splitting off bind_port() call into bind_port_helper(), with
635 * Revision 1.68 2002/03/04 20:17:32 oes
638 * Revision 1.67 2002/03/04 18:18:57 oes
639 * - Removed _DEBUG mode
640 * - Cleand up cmdline parsing
641 * - Introduced --no-daemon, --pidfile options
642 * - Cleaned up signal handling:
643 * - Terminate cleanly on INT, TERM and ABRT
644 * - Schedule logfile for re-opening on HUP
645 * - Ignore CHLD and PIPE
646 * - Leave the rest with their default handlers
647 * - Uniform handler registration
648 * - Added usage() function
649 * - Played styleguide police
651 * Revision 1.66 2002/03/03 15:06:55 oes
652 * Re-enabled automatic config reloading
654 * Revision 1.65 2002/03/03 14:49:11 oes
655 * Fixed CLF logging: Now uses client's original HTTP request
657 * Revision 1.64 2002/03/03 09:18:03 joergs
658 * Made jumbjuster work on AmigaOS again.
660 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
661 * Clean up a little CRLF unpleasantness that suddenly appeared
663 * Revision 1.62 2002/02/20 23:17:23 jongfoster
664 * Detecting some out-of memory conditions and exiting with a log message.
666 * Revision 1.61 2002/01/17 21:01:52 jongfoster
667 * Moving all our URL and URL pattern parsing code to urlmatch.c.
669 * Revision 1.60 2001/12/30 14:07:32 steudten
670 * - Add signal handling (unix)
671 * - Add SIGHUP handler (unix)
672 * - Add creation of pidfile (unix)
673 * - Add action 'top' in rc file (RH)
674 * - Add entry 'SIGNALS' to manpage
675 * - Add exit message to logfile (unix)
677 * Revision 1.59 2001/12/13 14:07:18 oes
678 * Fixed Bug: 503 error page now sent OK
680 * Revision 1.58 2001/11/30 23:37:24 jongfoster
681 * Renaming the Win32 config file to config.txt - this is almost the
682 * same as the corresponding UNIX name "config"
684 * Revision 1.57 2001/11/16 00:47:43 jongfoster
685 * Changing the tty-disconnection code to use setsid().
687 * Revision 1.56 2001/11/13 20:20:54 jongfoster
688 * Tabs->spaces, fixing a bug with missing {} around an if()
690 * Revision 1.55 2001/11/13 20:14:53 jongfoster
691 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
693 * Revision 1.54 2001/11/07 00:03:14 steudten
694 * Give reliable return value if an error
695 * occurs not just 0 with new daemon mode.
697 * Revision 1.53 2001/11/05 21:41:43 steudten
698 * Add changes to be a real daemon just for unix os.
699 * (change cwd to /, detach from controlling tty, set
700 * process group and session leader to the own process.
702 * Add some fatal-error log message for failed malloc().
703 * Add '-d' if compiled with 'configure --with-debug' to
704 * enable debug output.
706 * Revision 1.52 2001/10/26 20:11:20 jongfoster
707 * Fixing type mismatch
709 * Revision 1.51 2001/10/26 17:38:28 oes
712 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
713 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
714 * threads to call select() simultaneously. So, it's time to do a real, live,
715 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
716 * (native). Both versions will work, but using __OS2__ offers multi-threading.
718 * Revision 1.49 2001/10/23 21:41:35 jongfoster
719 * Added call to initialize the (statically-allocated of course)
720 * "out of memory" CGI response.
722 * Revision 1.48 2001/10/10 19:56:46 jongfoster
723 * Moving some code that wasn't cookie-related out of an #ifdef
726 * Revision 1.47 2001/10/10 16:44:36 oes
727 * Added CONNECT destination port limitation check
729 * Revision 1.46 2001/10/08 15:17:41 oes
730 * Re-enabled SSL forwarding
732 * Revision 1.45 2001/10/07 15:42:11 oes
733 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
735 * Moved downgrading of the HTTP version from parse_http_request to
736 * chat(), since we can't decide if it is necessary before we have
737 * determined the actions for the URL. The HTTP command is now
738 * *always* re-built so the repairs need no longer be special-cased.
740 * filter_popups now gets a csp pointer so it can raise the new
741 * CSP_FLAG_MODIFIED flag.
745 * Added configurable size limit for the IOB. If the IOB grows so
746 * large that the next read would exceed the limit, the header
747 * is generated, and the header & unfiltered buffer are flushed
748 * to the client. Chat then continues in non-buffering,
749 * non-filtering body mode.
751 * Revision 1.44 2001/10/02 18:13:57 oes
754 * Revision 1.43 2001/10/02 15:32:13 oes
755 * Moved generation of hdr
757 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
758 * Cleaning up 2 compiler warnings on OS/2.
760 * Revision 1.41 2001/09/16 17:05:14 jongfoster
761 * Removing unused #include showarg.h
763 * Revision 1.40 2001/09/16 15:41:45 jongfoster
764 * Fixing signed/unsigned comparison warning.
766 * Revision 1.39 2001/09/16 13:21:27 jongfoster
767 * Changes to use new list functions.
769 * Revision 1.38 2001/09/16 13:01:46 jongfoster
770 * Removing redundant function call that zeroed zalloc()'d memory.
772 * Revision 1.37 2001/09/10 11:12:24 oes
773 * Deleted unused variable
775 * Revision 1.36 2001/09/10 10:56:15 oes
776 * Silenced compiler warnings
778 * Revision 1.35 2001/07/31 14:44:22 oes
779 * Deleted unused size parameter from filter_popups()
781 * Revision 1.34 2001/07/30 22:08:36 jongfoster
782 * Tidying up #defines:
783 * - All feature #defines are now of the form FEATURE_xxx
784 * - Permanently turned off WIN_GUI_EDIT
785 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
787 * Revision 1.33 2001/07/29 19:32:00 jongfoster
788 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
790 * Revision 1.32 2001/07/29 18:47:05 jongfoster
791 * Adding missing #include "loadcfg.h"
793 * Revision 1.31 2001/07/29 12:17:48 oes
794 * Applied pthread fix by Paul Lieverse
796 * Revision 1.30 2001/07/25 22:57:13 jongfoster
797 * __BEOS__ no longer overrides FEATURE_PTHREAD.
798 * This is because FEATURE_PTHREAD will soon be widely used, so I
799 * want to keep it simple.
801 * Revision 1.29 2001/07/24 12:47:06 oes
802 * Applied BeOS support update by Eugenia
804 * Revision 1.28 2001/07/23 13:26:12 oes
805 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
807 * Revision 1.27 2001/07/19 19:09:47 haroon
808 * - Added code to take care of the situation where while processing the first
809 * server response (which includes the server header), after finding the end
810 * of the headers we were not looking past the end of the headers for
811 * content modification. I enabled it for filter_popups.
812 * Someone else should look to see if other similar operations should be
813 * done to the discarded portion of the buffer.
815 * Note 2001/07/20: No, the other content modification mechanisms will process
816 * the whole iob later anyway. --oes
818 * Revision 1.26 2001/07/18 12:31:36 oes
821 * Revision 1.25 2001/07/15 19:43:49 jongfoster
822 * Supports POSIX threads.
823 * Also removed some unused #includes.
825 * Revision 1.24 2001/07/13 14:00:40 oes
826 * - Generic content modification scheme:
827 * Each feature has its own applicability flag that is set
828 * from csp->action->flags.
829 * Replaced the "filtering" int flag , by a function pointer
830 * "content_filter" to the function that will do the content
831 * modification. If it is != NULL, the document will be buffered
832 * and processed through *content_filter, which must set
833 * csp->content_length and return a modified copy of the body
834 * or return NULL (on failiure).
835 * - Changed csp->is_text to the more generic bitmap csp->content_type
836 * which can currently take the valued CT_TEXT or CT_GIF
838 * - Removed all #ifdef PCRS
840 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
841 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
842 * the `block_acl' call. This prevents a compilation error when the user
843 * does not wish to use the "ACL" feature.
845 * Revision 1.22 2001/06/29 21:45:41 oes
846 * Indentation, CRLF->LF, Tab-> Space
848 * Revision 1.21 2001/06/29 13:29:36 oes
849 * - Cleaned up, improved comments
850 * - Unified all possible interceptors (CGI,
851 * block, trust, fast_redirect) in one
852 * place, with one (CGI) answer generation
853 * mechansim. Much clearer now.
854 * - Removed the GIF image generation, which
855 * is now done in filters.c:block_url()
856 * - Made error conditions like domain lookup
857 * failiure or (various) problems while talking
858 * to the server use cgi.c:error_response()
859 * instead of generating HTML/HTTP in chat() (yuck!)
860 * - Removed logentry from cancelled commit
862 * Revision 1.20 2001/06/09 10:55:28 jongfoster
863 * Changing BUFSIZ ==> BUFFER_SIZE
865 * Revision 1.19 2001/06/07 23:12:52 jongfoster
866 * Replacing function pointer in struct gateway with a directly
867 * called function forwarded_connect().
868 * Replacing struct gateway with struct forward_spec
870 * Revision 1.18 2001/06/03 19:12:16 oes
871 * introduced new cgi handling
873 * Revision 1.17 2001/06/01 20:07:23 jongfoster
874 * Now uses action +image-blocker{} rather than config->tinygif
876 * Revision 1.16 2001/06/01 18:49:17 jongfoster
877 * Replaced "list_share" with "list" - the tiny memory gain was not
878 * worth the extra complexity.
880 * Revision 1.15 2001/05/31 21:24:47 jongfoster
881 * Changed "permission" to "action" throughout.
882 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
883 * Moved vanilla wafer check into chat(), since we must now
884 * decide whether or not to add it based on the URL.
886 * Revision 1.14 2001/05/29 20:14:01 joergs
887 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
890 * Revision 1.13 2001/05/29 09:50:24 jongfoster
891 * Unified blocklist/imagelist/permissionslist.
892 * File format is still under discussion, but the internal changes
895 * Also modified interceptor behaviour:
896 * - We now intercept all URLs beginning with one of the following
897 * prefixes (and *only* these prefixes):
899 * * http://ijbswa.sf.net/config/
900 * * http://ijbswa.sourceforge.net/config/
901 * - New interceptors "home page" - go to http://i.j.b/ to see it.
902 * - Internal changes so that intercepted and fast redirect pages
903 * are not replaced with an image.
904 * - Interceptors now have the option to send a binary page direct
905 * to the client. (i.e. ijb-send-banner uses this)
906 * - Implemented show-url-info interceptor. (Which is why I needed
907 * the above interceptors changes - a typical URL is
908 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
909 * The previous mechanism would not have intercepted that, and
910 * if it had been intercepted then it then it would have replaced
913 * Revision 1.12 2001/05/27 22:17:04 oes
915 * - re_process_buffer no longer writes the modified buffer
916 * to the client, which was very ugly. It now returns the
917 * buffer, which it is then written by chat.
919 * - content_length now adjusts the Content-Length: header
920 * for modified documents rather than crunch()ing it.
921 * (Length info in csp->content_length, which is 0 for
922 * unmodified documents)
924 * - For this to work, sed() is called twice when filtering.
926 * Revision 1.11 2001/05/26 17:27:53 jongfoster
927 * Added support for CLF and fixed LOG_LEVEL_LOG.
928 * Also did CRLF->LF fix of my previous patch.
930 * Revision 1.10 2001/05/26 15:26:15 jongfoster
931 * ACL feature now provides more security by immediately dropping
932 * connections from untrusted hosts.
934 * Revision 1.9 2001/05/26 00:28:36 jongfoster
935 * Automatic reloading of config file.
936 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
937 * Most of the global variables have been moved to a new
938 * struct configuration_spec, accessed through csp->config->globalname
939 * Most of the globals remaining are used by the Win32 GUI.
941 * Revision 1.8 2001/05/25 22:43:18 jongfoster
942 * Fixing minor memory leak and buffer overflow.
944 * Revision 1.7 2001/05/25 22:34:30 jongfoster
947 * Revision 1.6 2001/05/23 00:13:58 joergs
948 * AmigaOS support fixed.
950 * Revision 1.5 2001/05/22 18:46:04 oes
952 * - Enabled filtering banners by size rather than URL
953 * by adding patterns that replace all standard banner
954 * sizes with the "Junkbuster" gif to the re_filterfile
956 * - Enabled filtering WebBugs by providing a pattern
957 * which kills all 1x1 images
959 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
960 * which is selected by the (nonstandard and therefore
961 * capital) letter 'U' in the option string.
962 * It causes the quantifiers to be ungreedy by default.
963 * Appending a ? turns back to greedy (!).
965 * - Added a new interceptor ijb-send-banner, which
966 * sends back the "Junkbuster" gif. Without imagelist or
967 * MSIE detection support, or if tinygif = 1, or the
968 * URL isn't recognized as an imageurl, a lame HTML
969 * explanation is sent instead.
971 * - Added new feature, which permits blocking remote
972 * script redirects and firing back a local redirect
974 * The feature is conditionally compiled, i.e. it
975 * can be disabled with --disable-fast-redirects,
976 * plus it must be activated by a "fast-redirects"
977 * line in the config file, has its own log level
978 * and of course wants to be displayed by show-proxy-args
979 * Note: Boy, all the #ifdefs in 1001 locations and
980 * all the fumbling with configure.in and acconfig.h
981 * were *way* more work than the feature itself :-(
983 * - Because a generic redirect template was needed for
984 * this, tinygif = 3 now uses the same.
986 * - Moved GIFs, and other static HTTP response templates
991 * - Removed some >400 CRs again (Jon, you really worked
994 * Revision 1.4 2001/05/21 19:34:01 jongfoster
995 * Made failure to bind() a fatal error.
997 * Revision 1.3 2001/05/20 01:21:20 jongfoster
998 * Version 2.9.4 checkin.
999 * - Merged popupfile and cookiefile, and added control over PCRS
1000 * filtering, in new "permissionsfile".
1001 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1002 * file error you now get a message box (in the Win32 GUI) rather
1003 * than the program exiting with no explanation.
1004 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1006 * - Removed tabs from "config"
1007 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1008 * - Bumped up version number.
1010 * Revision 1.2 2001/05/17 22:34:44 oes
1011 * - Added hint on GIF char array generation to jcc.c
1012 * - Cleaned CRLF's from the sources and related files
1013 * - Repaired logging for REF and FRC
1015 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1016 * Initial import of version 2.9.3 source tree
1019 *********************************************************************/
1025 #include <sys/types.h>
1034 # ifndef FEATURE_PTHREAD
1038 # include <windows.h>
1039 # include <process.h>
1040 # endif /* ndef FEATURE_PTHREAD */
1043 # ifndef _WIN_CONSOLE
1044 # include "w32log.h"
1045 # endif /* ndef _WIN_CONSOLE */
1046 # include "w32svrapi.h"
1048 #else /* ifndef _WIN32 */
1050 # if !defined (__OS2__)
1051 # include <unistd.h>
1052 # include <sys/wait.h>
1053 # endif /* ndef __OS2__ */
1054 # include <sys/time.h>
1055 # include <sys/stat.h>
1056 # include <sys/ioctl.h>
1059 #include <sys/termios.h>
1067 # include <signal.h>
1070 # include <socket.h> /* BeOS has select() for sockets only. */
1071 # include <OS.h> /* declarations for threads and stuff. */
1074 # if defined(__EMX__) || defined(__OS2__)
1075 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1080 #define bzero(B,N) memset(B,0x00,n)
1084 # include <select.h>
1089 #include "project.h"
1092 #include "filters.h"
1093 #include "loaders.h"
1094 #include "parsers.h"
1095 #include "miscutil.h"
1097 #include "jbsockets.h"
1098 #include "gateway.h"
1099 #include "actions.h"
1101 #include "loadcfg.h"
1102 #include "urlmatch.h"
1104 const char jcc_h_rcs[] = JCC_H_VERSION;
1105 const char project_h_rcs[] = PROJECT_H_VERSION;
1108 struct client_state clients[1];
1109 struct file_list files[1];
1111 #ifdef FEATURE_STATISTICS
1112 int urls_read = 0; /* total nr of urls read inc rejected */
1113 int urls_rejected = 0; /* total nr of urls rejected */
1114 #endif /* def FEATURE_STATISTICS */
1116 #ifdef FEATURE_GRACEFUL_TERMINATION
1117 int g_terminate = 0;
1120 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1121 static void sig_handler(int the_signal);
1123 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1124 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1125 static jb_err get_server_headers(struct client_state *csp);
1126 static const char *crunch_reason(const struct http_response *rsp);
1127 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1129 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1131 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1132 static jb_err change_request_destination(struct client_state *csp);
1133 static void chat(struct client_state *csp);
1134 static void serve(struct client_state *csp);
1135 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1136 static void usage(const char *myname);
1138 static void initialize_mutexes(void);
1139 static jb_socket bind_port_helper(struct configuration_spec *config);
1140 static void listen_loop(void);
1143 void serve(struct client_state *csp);
1144 #else /* ifndef AMIGA */
1145 static void serve(struct client_state *csp);
1146 #endif /* def AMIGA */
1149 static int32 server_thread(void *data);
1150 #endif /* def __BEOS__ */
1153 #define sleep(N) Sleep(((N) * 1000))
1157 #define sleep(N) DosSleep(((N) * 100))
1160 #ifdef FEATURE_PTHREAD
1161 pthread_mutex_t log_mutex;
1162 pthread_mutex_t log_init_mutex;
1164 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1165 pthread_mutex_t resolver_mutex;
1166 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1168 #ifndef HAVE_GMTIME_R
1169 pthread_mutex_t gmtime_mutex;
1170 #endif /* ndef HAVE_GMTIME_R */
1172 #ifndef HAVE_LOCALTIME_R
1173 pthread_mutex_t localtime_mutex;
1174 #endif /* ndef HAVE_GMTIME_R */
1177 pthread_mutex_t rand_mutex;
1178 #endif /* ndef HAVE_RANDOM */
1180 #endif /* FEATURE_PTHREAD */
1182 #if defined(unix) || defined(__EMX__)
1183 const char *basedir = NULL;
1184 const char *pidfile = NULL;
1185 int received_hup_signal = 0;
1186 #endif /* defined unix */
1188 /* HTTP snipplets. */
1189 static const char CSUCCEED[] =
1190 "HTTP/1.0 200 Connection established\r\n"
1191 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1193 static const char CHEADER[] =
1194 "HTTP/1.0 400 Invalid header received from client\r\n"
1195 "Proxy-Agent: Privoxy " VERSION "\r\n"
1196 "Content-Type: text/plain\r\n"
1197 "Connection: close\r\n\r\n"
1198 "Invalid header received from client.\r\n";
1200 static const char FTP_RESPONSE[] =
1201 "HTTP/1.0 400 Invalid request received from client\r\n"
1202 "Content-Type: text/plain\r\n"
1203 "Connection: close\r\n\r\n"
1204 "Invalid request. Privoxy doesn't support FTP.\r\n";
1206 static const char GOPHER_RESPONSE[] =
1207 "HTTP/1.0 400 Invalid request received from client\r\n"
1208 "Content-Type: text/plain\r\n"
1209 "Connection: close\r\n\r\n"
1210 "Invalid request. Privoxy doesn't support gopher.\r\n";
1212 /* XXX: should be a template */
1213 static const char MISSING_DESTINATION_RESPONSE[] =
1214 "HTTP/1.0 400 Bad request received from client\r\n"
1215 "Proxy-Agent: Privoxy " VERSION "\r\n"
1216 "Content-Type: text/plain\r\n"
1217 "Connection: close\r\n\r\n"
1218 "Bad request. Privoxy was unable to extract the destination.\r\n";
1220 /* XXX: should be a template */
1221 static const char NO_SERVER_DATA_RESPONSE[] =
1222 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1223 "Proxy-Agent: Privoxy " VERSION "\r\n"
1224 "Content-Type: text/plain\r\n"
1225 "Connection: close\r\n\r\n"
1226 "Empty server or forwarder response.\r\n"
1227 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1229 /* XXX: should be a template */
1230 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1231 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1232 "Proxy-Agent: Privoxy " VERSION "\r\n"
1233 "Content-Type: text/plain\r\n"
1234 "Connection: close\r\n\r\n"
1235 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1238 /* XXX: should be a template */
1239 static const char NULL_BYTE_RESPONSE[] =
1240 "HTTP/1.0 400 Bad request received from client\r\n"
1241 "Proxy-Agent: Privoxy " VERSION "\r\n"
1242 "Content-Type: text/plain\r\n"
1243 "Connection: close\r\n\r\n"
1244 "Bad request. Null byte(s) before end of request.\r\n";
1247 /* XXX: should be a template */
1248 static const char MESSED_UP_REQUEST_RESPONSE[] =
1249 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1250 "Proxy-Agent: Privoxy " VERSION "\r\n"
1251 "Content-Type: text/plain\r\n"
1252 "Connection: close\r\n\r\n"
1253 "Bad request. Messed up with header filters.\r\n";
1255 /* A function to crunch a response */
1256 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1258 /* Crunch function flags */
1259 #define CF_NO_FLAGS 0
1260 /* Cruncher applies to forced requests as well */
1261 #define CF_IGNORE_FORCE 1
1262 /* Crunched requests are counted for the block statistics */
1263 #define CF_COUNT_AS_REJECT 2
1265 /* A crunch function and its flags */
1268 const crunch_func_ptr cruncher;
1272 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1274 /* Complete list of cruncher functions */
1275 static const struct cruncher crunchers_all[] = {
1276 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1277 { block_url, CF_COUNT_AS_REJECT },
1278 #ifdef FEATURE_TRUST
1279 { trust_url, CF_COUNT_AS_REJECT },
1280 #endif /* def FEATURE_TRUST */
1281 { redirect_url, CF_NO_FLAGS },
1282 { dispatch_cgi, CF_IGNORE_FORCE},
1286 /* Light version, used after tags are applied */
1287 static const struct cruncher crunchers_light[] = {
1288 { block_url, CF_COUNT_AS_REJECT },
1289 { redirect_url, CF_NO_FLAGS },
1294 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1295 /*********************************************************************
1297 * Function : sig_handler
1299 * Description : Signal handler for different signals.
1300 * Exit gracefully on TERM and INT
1301 * or set a flag that will cause the errlog
1302 * to be reopened by the main thread on HUP.
1305 * 1 : the_signal = the signal cause this function to call
1309 *********************************************************************/
1310 static void sig_handler(int the_signal)
1316 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1327 received_hup_signal = 1;
1332 * We shouldn't be here, unless we catch signals
1333 * in main() that we can't handle here!
1335 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1343 /*********************************************************************
1345 * Function : client_protocol_is_unsupported
1347 * Description : Checks if the client used a known unsupported
1348 * protocol and deals with it by sending an error
1352 * 1 : csp = Current client state (buffers, headers, etc...)
1353 * 2 : req = the first request line send by the client
1355 * Returns : TRUE if an error response has been generated, or
1356 * FALSE if the request doesn't look invalid.
1358 *********************************************************************/
1359 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1362 * If it's a FTP or gopher request, we don't support it.
1364 * These checks are better than nothing, but they might
1365 * not work in all configurations and some clients might
1366 * have problems digesting the answer.
1368 * They should, however, never cause more problems than
1369 * Privoxy's old behaviour (returning the misleading HTML
1372 * "Could not resolve http://(ftp|gopher)://example.org").
1374 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1376 const char *response = NULL;
1377 const char *protocol = NULL;
1379 if (!strncmpic(req, "GET ftp://", 10))
1381 response = FTP_RESPONSE;
1386 response = GOPHER_RESPONSE;
1387 protocol = "GOPHER";
1389 log_error(LOG_LEVEL_ERROR,
1390 "%s tried to use Privoxy as %s proxy: %s",
1391 csp->ip_addr_str, protocol, req);
1392 log_error(LOG_LEVEL_CLF,
1393 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1395 write_socket(csp->cfd, response, strlen(response));
1404 /*********************************************************************
1406 * Function : get_request_destination_elsewhere
1408 * Description : If the client's request was redirected into
1409 * Privoxy without the client's knowledge,
1410 * the request line lacks the destination host.
1412 * This function tries to get it elsewhere,
1413 * provided accept-intercepted-requests is enabled.
1415 * "Elsewhere" currently only means "Host: header",
1416 * but in the future we may ask the redirecting
1417 * packet filter to look the destination up.
1419 * If the destination stays unknown, an error
1420 * response is send to the client and headers
1421 * are freed so that chat() can return directly.
1424 * 1 : csp = Current client state (buffers, headers, etc...)
1425 * 2 : headers = a header list
1427 * Returns : JB_ERR_OK if the destination is now known, or
1428 * JB_ERR_PARSE if it isn't.
1430 *********************************************************************/
1431 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1435 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1437 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1438 " Privoxy isn't configured to accept intercepted requests.",
1439 csp->ip_addr_str, csp->http->cmd);
1440 /* XXX: Use correct size */
1441 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1442 csp->ip_addr_str, csp->http->cmd);
1444 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1445 destroy_list(headers);
1447 return JB_ERR_PARSE;
1449 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1451 /* Split the domain we just got for pattern matching */
1452 init_domain_components(csp->http);
1458 /* We can't work without destination. Go spread the news.*/
1460 req = list_to_text(headers);
1462 /* XXX: Use correct size */
1463 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1464 csp->ip_addr_str, csp->http->cmd);
1465 log_error(LOG_LEVEL_ERROR,
1466 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1467 csp->ip_addr_str, csp->http->cmd, req);
1470 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1471 destroy_list(headers);
1473 return JB_ERR_PARSE;
1476 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1477 * to get the destination IP address, use it as host directly
1478 * or do a reverse DNS lookup first.
1483 /*********************************************************************
1485 * Function : get_server_headers
1487 * Description : Parses server headers in iob and fills them
1488 * into csp->headers so that they can later be
1492 * 1 : csp = Current client state (buffers, headers, etc...)
1494 * Returns : JB_ERR_OK if everything went fine, or
1495 * JB_ERR_PARSE if the headers were incomplete.
1497 *********************************************************************/
1498 static jb_err get_server_headers(struct client_state *csp)
1500 int continue_hack_in_da_house = 0;
1503 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1508 * continue hack in da house. Ignore the ending of
1509 * this head and continue enlisting header lines.
1510 * The reason is described below.
1512 enlist(csp->headers, "");
1513 continue_hack_in_da_house = 0;
1516 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1519 * It's a bodyless continue response, don't
1520 * stop header parsing after reaching its end.
1522 * As a result Privoxy will concatenate the
1523 * next response's head and parse and deliver
1524 * the headers as if they belonged to one request.
1526 * The client will separate them because of the
1527 * empty line between them.
1529 * XXX: What we're doing here is clearly against
1530 * the intended purpose of the continue header,
1531 * and under some conditions (HTTP/1.0 client request)
1532 * it's a standard violation.
1534 * Anyway, "sort of against the spec" is preferable
1535 * to "always getting confused by Continue responses"
1536 * (Privoxy's behaviour before this hack was added)
1538 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1539 continue_hack_in_da_house = 1;
1541 else if (*header == '\0')
1544 * If the header is empty, but the Continue hack
1545 * isn't active, we can assume that we reached the
1546 * end of the buffer before we hit the end of the
1549 * Inform the caller an let it decide how to handle it.
1551 return JB_ERR_PARSE;
1554 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1557 * XXX: Should we quit the request and return a
1558 * out of memory error page instead?
1560 log_error(LOG_LEVEL_ERROR,
1561 "Out of memory while enlisting server headers. %s lost.",
1571 /*********************************************************************
1573 * Function : crunch_reason
1575 * Description : Translates the crunch reason code into a string.
1578 * 1 : rsp = a http_response
1580 * Returns : A string with the crunch reason or an error description.
1582 *********************************************************************/
1583 static const char *crunch_reason(const struct http_response *rsp)
1585 char * reason = NULL;
1587 assert(rsp != NULL);
1590 return "Internal error while searching for crunch reason";
1593 switch (rsp->reason)
1595 case RSP_REASON_UNSUPPORTED:
1596 reason = "Unsupported HTTP feature";
1598 case RSP_REASON_BLOCKED:
1601 case RSP_REASON_UNTRUSTED:
1602 reason = "Untrusted";
1604 case RSP_REASON_REDIRECTED:
1605 reason = "Redirected";
1607 case RSP_REASON_CGI_CALL:
1608 reason = "CGI Call";
1610 case RSP_REASON_NO_SUCH_DOMAIN:
1611 reason = "DNS failure";
1613 case RSP_REASON_FORWARDING_FAILED:
1614 reason = "Forwarding failed";
1616 case RSP_REASON_CONNECT_FAILED:
1617 reason = "Connection failure";
1619 case RSP_REASON_OUT_OF_MEMORY:
1620 reason = "Out of memory (may mask other reasons)";
1623 reason = "No reason recorded";
1631 /*********************************************************************
1633 * Function : send_crunch_response
1635 * Description : Delivers already prepared response for
1636 * intercepted requests, logs the interception
1637 * and frees the response.
1640 * 1 : csp = Current client state (buffers, headers, etc...)
1641 * 1 : rsp = Fully prepared response. Will be freed on exit.
1643 * Returns : Nothing.
1645 *********************************************************************/
1646 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1648 const struct http_request *http = csp->http;
1649 char status_code[4];
1651 assert(rsp != NULL);
1652 assert(rsp->head != NULL);
1657 * Not supposed to happen. If it does
1658 * anyway, treat it as an unknown error.
1660 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1661 /* return code doesn't matter */
1666 /* If rsp is still NULL, we have serious internal problems. */
1667 log_error(LOG_LEVEL_FATAL,
1668 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1672 * Extract the status code from the actual head
1673 * that was send to the client. It is the only
1674 * way to get it right for all requests, including
1675 * the fixed ones for out-of-memory problems.
1677 * A head starts like this: 'HTTP/1.1 200...'
1681 status_code[0] = rsp->head[9];
1682 status_code[1] = rsp->head[10];
1683 status_code[2] = rsp->head[11];
1684 status_code[3] = '\0';
1686 /* Write the answer to the client */
1687 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1688 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1690 /* There is nothing we can do about it. */
1691 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1694 /* Log that the request was crunched and why. */
1695 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1696 http->hostport, http->path, crunch_reason(rsp));
1697 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1698 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1700 /* Clean up and return */
1701 if (cgi_error_memory() != rsp)
1703 free_http_response(rsp);
1710 /*********************************************************************
1712 * Function : request_contains_null_bytes
1714 * Description : Checks for NULL bytes in the request and sends
1715 * an error message to the client if any were found.
1717 * XXX: currently not used, see comment in chat().
1720 * 1 : csp = Current client state (buffers, headers, etc...)
1721 * 2 : buf = Data from the client's request to check.
1722 * 3 : len = The data length.
1724 * Returns : TRUE if the request contained one or more NULL bytes, or
1727 *********************************************************************/
1728 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1730 size_t c_len; /* Request lenght when treated as C string */
1732 c_len = strlen(buf);
1737 * Null byte(s) found. Log the request,
1738 * return an error response and hang up.
1740 size_t tmp_len = c_len;
1745 * Replace NULL byte(s) with '°' characters
1746 * so the request can be logged as string.
1747 * XXX: Is there a better replacement character?
1750 tmp_len += strlen(buf+tmp_len);
1751 } while (tmp_len < len);
1753 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1754 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1755 log_error(LOG_LEVEL_HEADER,
1756 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1758 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1760 /* XXX: Log correct size */
1761 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1771 /*********************************************************************
1773 * Function : crunch_response_triggered
1775 * Description : Checks if the request has to be crunched,
1776 * and delivers the crunch response if necessary.
1779 * 1 : csp = Current client state (buffers, headers, etc...)
1780 * 2 : crunchers = list of cruncher functions to run
1782 * Returns : TRUE if the request was answered with a crunch response
1785 *********************************************************************/
1786 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1788 struct http_response *rsp = NULL;
1789 const struct cruncher *c;
1792 * If CGI request crunching is disabled,
1793 * check the CGI dispatcher out of order to
1794 * prevent unintentional blocks or redirects.
1796 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1797 && (NULL != (rsp = dispatch_cgi(csp))))
1799 /* Deliver, log and free the interception response. */
1800 send_crunch_response(csp, rsp);
1804 for (c = crunchers; c->cruncher != NULL; c++)
1807 * Check the cruncher if either Privoxy is toggled
1808 * on and the request isn't forced, or if the cruncher
1809 * applies to forced requests as well.
1811 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1812 !(csp->flags & CSP_FLAG_FORCED)) ||
1813 (c->flags & CF_IGNORE_FORCE))
1815 rsp = c->cruncher(csp);
1818 /* Deliver, log and free the interception response. */
1819 send_crunch_response(csp, rsp);
1820 #ifdef FEATURE_STATISTICS
1821 if (c->flags & CF_COUNT_AS_REJECT)
1823 csp->flags |= CSP_FLAG_REJECTED;
1825 #endif /* def FEATURE_STATISTICS */
1836 /*********************************************************************
1838 * Function : build_request_line
1840 * Description : Builds the HTTP request line.
1842 * If a HTTP forwarder is used it expects the whole URL,
1843 * web servers only get the path.
1846 * 1 : csp = Current client state (buffers, headers, etc...)
1847 * 2 : fwd = The forwarding spec used for the request
1848 * XXX: Should use http->fwd instead.
1849 * 3 : request_line = The old request line which will be replaced.
1851 * Returns : Nothing. Terminates in case of memory problems.
1853 *********************************************************************/
1854 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1856 struct http_request *http = csp->http;
1858 assert(http->ssl == 0);
1861 * Downgrade http version from 1.1 to 1.0
1862 * if +downgrade action applies.
1864 if ( (csp->action->flags & ACTION_DOWNGRADE)
1865 && (!strcmpic(http->ver, "HTTP/1.1")))
1868 http->ver = strdup("HTTP/1.0");
1870 if (http->ver == NULL)
1872 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1877 * Rebuild the request line.
1879 freez(*request_line);
1880 *request_line = strdup(http->gpc);
1881 string_append(request_line, " ");
1883 if (fwd->forward_host)
1885 string_append(request_line, http->url);
1889 string_append(request_line, http->path);
1891 string_append(request_line, " ");
1892 string_append(request_line, http->ver);
1894 if (*request_line == NULL)
1896 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1898 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1902 /*********************************************************************
1904 * Function : change_request_destination
1906 * Description : Parse a (rewritten) request line and regenerate
1907 * the http request data.
1910 * 1 : csp = Current client state (buffers, headers, etc...)
1912 * Returns : Forwards the parse_http_request() return code.
1913 * Terminates in case of memory problems.
1915 *********************************************************************/
1916 static jb_err change_request_destination(struct client_state *csp)
1918 struct http_request *http = csp->http;
1921 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1922 free_http_request(http);
1923 err = parse_http_request(csp->headers->first->str, http, csp);
1924 if (JB_ERR_OK != err)
1926 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1927 jb_err_to_string(err));
1931 /* XXX: ocmd is a misleading name */
1932 http->ocmd = strdup(http->cmd);
1933 if (http->ocmd == NULL)
1935 log_error(LOG_LEVEL_FATAL,
1936 "Out of memory copying rewritten HTTP request line");
1944 /*********************************************************************
1948 * Description : Once a connection to the client has been accepted,
1949 * this function is called (via serve()) to handle the
1950 * main business of the communication. When this
1951 * function returns, the caller must close the client
1954 * FIXME: chat is nearly thousand lines long.
1958 * 1 : csp = Current client state (buffers, headers, etc...)
1960 * Returns : Nothing.
1962 *********************************************************************/
1963 static void chat(struct client_state *csp)
1965 char buf[BUFFER_SIZE];
1973 int ms_iis5_hack = 0;
1974 size_t byte_count = 0;
1975 int forwarded_connect_retries = 0;
1976 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1977 const struct forward_spec * fwd;
1978 struct http_request *http;
1979 int len; /* for buffer sizes (and negative error codes) */
1982 /* Function that does the content filtering for the current request */
1983 filter_function_ptr content_filter = NULL;
1985 /* Skeleton for HTTP response, if we should intercept the request */
1986 struct http_response *rsp;
1988 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1989 struct list header_list;
1990 struct list *headers = &header_list;
1994 memset(buf, 0, sizeof(buf));
1997 * Read the client's request. Note that since we're not using select() we
1998 * could get blocked here if a client connected, then didn't say anything!
2003 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2005 if (len <= 0) break; /* error! */
2008 * If there is no memory left for buffering the
2009 * request, there is nothing we can do but hang up
2011 if (add_to_iob(csp, buf, len))
2016 req = get_header(csp->iob);
2018 } while ((NULL != req) && ('\0' == *req));
2020 if ((NULL != req) && ('\0' != *req))
2022 /* Request received. Validate and parse it. */
2026 * XXX: Temporary disabled to prevent problems
2027 * with POST requests whose bodies are allowed to
2028 * contain NULL bytes. BR#1730105.
2030 * The main purpose of this check is to properly
2031 * log stuff like BitTorrent traffic and other junk
2032 * that hits public proxies. It's not required for
2033 * Privoxy to functions as those requests are discarded
2036 * It probably should be rewritten to only check
2037 * the head of the request. Another option would
2038 * be to let all POST requests pass, although that
2039 * may not be good enough.
2041 if (request_contains_null_bytes(csp, buf, len))
2043 /* NULL bytes found and dealt with, just hang up. */
2048 /* Does the request line look invalid? */
2049 if (client_protocol_is_unsupported(csp, req))
2052 * Yes. The request has already been
2053 * answered with a error response, the buffers
2054 * were freed and we're done with chatting.
2059 #ifdef FEATURE_FORCE_LOAD
2061 * If this request contains the FORCE_PREFIX and blocks
2062 * aren't enforced, get rid of it and set the force flag.
2064 if (strstr(req, FORCE_PREFIX))
2066 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2068 log_error(LOG_LEVEL_FORCE,
2069 "Ignored force prefix in request: \"%s\".", req);
2073 strclean(req, FORCE_PREFIX);
2074 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2075 csp->flags |= CSP_FLAG_FORCED;
2079 #endif /* def FEATURE_FORCE_LOAD */
2080 err = parse_http_request(req, http, csp);
2081 if (JB_ERR_OK != err)
2083 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2089 if (http->cmd == NULL)
2091 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2092 /* XXX: Use correct size */
2093 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2094 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2096 free_http_request(http);
2100 /* grab the rest of the client's headers */
2104 p = get_header(csp->iob);
2108 /* There are no additional headers to read. */
2115 * We didn't receive a complete header
2116 * line yet, get the rest of it.
2118 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2121 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2122 destroy_list(headers);
2126 if (add_to_iob(csp, buf, len))
2129 * If there is no memory left for buffering the
2130 * request, there is nothing we can do but hang up
2132 destroy_list(headers);
2139 * We were able to read a complete
2140 * header and can finaly enlist it.
2147 if (http->host == NULL)
2150 * If we still don't know the request destination,
2151 * the request is invalid or the client uses
2152 * Privoxy without its knowledge.
2154 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2157 * Our attempts to get the request destination
2158 * elsewhere failed or Privoxy is configured
2159 * to only accept proxy requests.
2161 * An error response has already been send
2162 * and we're done here.
2169 * Determine the actions for this URL
2171 #ifdef FEATURE_TOGGLE
2172 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2174 /* Most compatible set of actions (i.e. none) */
2175 init_current_action(csp->action);
2178 #endif /* ndef FEATURE_TOGGLE */
2180 get_url_actions(csp, http);
2184 * Save a copy of the original request for logging
2186 http->ocmd = strdup(http->cmd);
2188 if (http->ocmd == NULL)
2190 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2193 enlist(csp->headers, http->cmd);
2195 /* Append the previously read headers */
2196 list_append_list_unique(csp->headers, headers);
2197 destroy_list(headers);
2199 err = sed(csp, FILTER_CLIENT_HEADERS);
2200 if (JB_ERR_OK != err)
2202 assert(err == JB_ERR_PARSE);
2203 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2205 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2207 /* Check request line for rewrites. */
2208 if ((NULL == csp->headers->first->str)
2209 || (strcmp(http->cmd, csp->headers->first->str) &&
2210 (JB_ERR_OK != change_request_destination(csp))))
2213 * A header filter broke the request line - bail out.
2215 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2216 /* XXX: Use correct size */
2217 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2218 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2220 free_http_request(http);
2224 /* decide how to route the HTTP request */
2225 fwd = forward_url(csp, http);
2228 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2229 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2233 * build the http request to send to the server
2234 * we have to do one of the following:
2236 * create = use the original HTTP request to create a new
2237 * HTTP request that has either the path component
2238 * without the http://domainspec (w/path) or the
2239 * full orininal URL (w/url)
2240 * Note that the path and/or the HTTP version may
2241 * have been altered by now.
2243 * connect = Open a socket to the host:port of the server
2244 * and short-circuit server and client socket.
2246 * pass = Pass the request unchanged if forwarding a CONNECT
2247 * request to a parent proxy. Note that we'll be sending
2248 * the CFAIL message ourselves if connecting to the parent
2249 * fails, but we won't send a CSUCCEED message if it works,
2250 * since that would result in a double message (ours and the
2251 * parent's). After sending the request to the parent, we simply
2254 * here's the matrix:
2257 * +--------+--------+
2259 * 0 | create | connect|
2261 * Forwarding +--------+--------+
2263 * 1 | create | pass |
2265 * +--------+--------+
2269 if (http->ssl && connect_port_is_forbidden(csp))
2271 const char *acceptable_connect_ports =
2272 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2273 assert(NULL != acceptable_connect_ports);
2274 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2275 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2276 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2277 csp->action->flags |= ACTION_BLOCK;
2283 freez(csp->headers->first->str);
2284 build_request_line(csp, fwd, &csp->headers->first->str);
2287 hdr = list_to_text(csp->headers);
2290 /* FIXME Should handle error properly */
2291 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2295 * We have a request. Check if one of the crunchers wants it.
2297 if (crunch_response_triggered(csp, crunchers_all))
2300 * Yes. The client got the crunch response
2301 * and we are done here after cleaning up.
2304 list_remove_all(csp->headers);
2310 * The headers can't be removed earlier because
2311 * they were still needed for the referrer check
2312 * in case of CGI crunches.
2314 * XXX: Would it be worth to move the referrer check
2315 * into client_referrer() and set a flag if it's trusted?
2317 list_remove_all(csp->headers);
2319 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2321 if (fwd->forward_host)
2323 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2324 fwd->forward_host, fwd->forward_port, http->hostport);
2328 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2331 /* here we connect to the server, gateway, or the forwarder */
2333 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2334 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2336 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2337 forwarded_connect_retries, http->hostport);
2340 if (csp->sfd == JB_INVALID_SOCKET)
2342 if (fwd->type != SOCKS_NONE)
2345 rsp = error_response(csp, "forwarding-failed", errno);
2347 else if (errno == EINVAL)
2349 rsp = error_response(csp, "no-such-domain", errno);
2353 rsp = error_response(csp, "connect-failed", errno);
2354 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2359 /* Write the answer to the client */
2362 send_crunch_response(csp, rsp);
2369 if (fwd->forward_host || (http->ssl == 0))
2371 /* write the client's (modified) header to the server
2372 * (along with anything else that may be in the buffer)
2375 if (write_socket(csp->sfd, hdr, strlen(hdr))
2376 || (flush_socket(csp->sfd, csp->iob) < 0))
2378 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2381 rsp = error_response(csp, "connect-failed", errno);
2385 send_crunch_response(csp, rsp);
2395 * We're running an SSL tunnel and we're not forwarding,
2396 * so just send the "connect succeeded" message to the
2397 * client, flush the rest, and get out of the way.
2399 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2407 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2409 /* we're finished with the client's header */
2412 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2414 /* pass data between the client and server
2415 * until one or the other shuts down the connection.
2424 * FD_ZERO here seems to point to an errant macro which crashes.
2425 * So do this by hand for now...
2427 memset(&rfds,0x00,sizeof(fd_set));
2431 FD_SET(csp->cfd, &rfds);
2432 FD_SET(csp->sfd, &rfds);
2434 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2438 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2442 /* this is the body of the browser's request
2443 * just read it and write it.
2446 if (FD_ISSET(csp->cfd, &rfds))
2448 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2452 break; /* "game over, man" */
2455 if (write_socket(csp->sfd, buf, (size_t)len))
2457 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2464 * The server wants to talk. It could be the header or the body.
2465 * If `hdr' is null, then it's the header otherwise it's the body.
2466 * FIXME: Does `hdr' really mean `host'? No.
2470 if (FD_ISSET(csp->sfd, &rfds))
2473 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2477 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2479 if (http->ssl && (fwd->forward_host == NULL))
2482 * Just hang up. We already confirmed the client's CONNECT
2483 * request with status code 200 and unencrypted content is
2484 * no longer welcome.
2486 log_error(LOG_LEVEL_ERROR,
2487 "CONNECT already confirmed. Unable to tell the client about the problem.");
2490 else if (byte_count)
2493 * Just hang up. We already transmitted the original headers
2494 * and parts of the original content and therefore missed the
2495 * chance to send an error message (without risking data corruption).
2497 * XXX: we could retry with a fancy range request here.
2499 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2500 "Unable to tell the client about the problem.");
2504 rsp = error_response(csp, "connect-failed", errno);
2508 send_crunch_response(csp, rsp);
2514 /* Add a trailing zero. This lets filter_popups
2515 * use string operations.
2519 /* Normally, this would indicate that we've read
2520 * as much as the server has sent us and we can
2521 * close the client connection. However, Microsoft
2522 * in its wisdom has released IIS/5 with a bug that
2523 * prevents it from sending the trailing \r\n in
2524 * a 302 redirect header (and possibly other headers).
2525 * To work around this if we've haven't parsed
2526 * a full header we'll append a trailing \r\n
2527 * and see if this now generates a valid one.
2529 * This hack shouldn't have any impacts. If we've
2530 * already transmitted the header or if this is a
2531 * SSL connection, then we won't bother with this
2532 * hack. So we only work on partially received
2533 * headers. If we append a \r\n and this still
2534 * doesn't generate a valid header, then we won't
2535 * transmit anything to the client.
2540 if (server_body || http->ssl)
2543 * If we have been buffering up the document,
2544 * now is the time to apply content modification
2545 * and send the result to the client.
2549 p = execute_content_filter(csp, content_filter);
2551 * If the content filter fails, use the original
2552 * buffer and length.
2553 * (see p != NULL ? p : csp->iob->cur below)
2557 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2560 if (JB_ERR_OK != update_server_headers(csp))
2562 log_error(LOG_LEVEL_FATAL,
2563 "Failed to update server headers. after filtering.");
2566 hdr = list_to_text(csp->headers);
2569 /* FIXME Should handle error properly */
2570 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2573 if (write_socket(csp->cfd, hdr, strlen(hdr))
2574 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2576 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2586 break; /* "game over, man" */
2590 * This is NOT the body, so
2591 * Let's pretend the server just sent us a blank line.
2593 snprintf(buf, sizeof(buf), "\r\n");
2594 len = (int)strlen(buf);
2597 * Now, let the normal header parsing algorithm below do its
2598 * job. If it fails, we'll exit instead of continuing.
2605 * If this is an SSL connection or we're in the body
2606 * of the server document, just write it to the client,
2607 * unless we need to buffer the body for later content-filtering
2610 if (server_body || http->ssl)
2615 * If there is no memory left for buffering the content, or the buffer limit
2616 * has been reached, switch to non-filtering mode, i.e. make & write the
2617 * header, flush the iob and buf, and get out of the way.
2619 if (add_to_iob(csp, buf, len))
2624 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2626 hdr = list_to_text(csp->headers);
2630 * Memory is too tight to even generate the header.
2631 * Send our static "Out-of-memory" page.
2633 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2634 rsp = cgi_error_memory();
2635 send_crunch_response(csp, rsp);
2639 hdrlen = strlen(hdr);
2641 if (write_socket(csp->cfd, hdr, hdrlen)
2642 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2643 || (write_socket(csp->cfd, buf, (size_t)len)))
2645 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2652 * Reset the byte_count to the amount of bytes
2653 * we just flushed. len will be added a few lines below,
2654 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2656 byte_count = (size_t)flushed;
2658 content_filter = NULL;
2664 if (write_socket(csp->cfd, buf, (size_t)len))
2666 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2670 byte_count += (size_t)len;
2675 /* we're still looking for the end of the
2676 * server's header ... (does that make header
2677 * parsing an "out of body experience" ?
2681 * buffer up the data we just read. If that fails,
2682 * there's little we can do but send our static
2683 * out-of-memory page.
2685 if (add_to_iob(csp, buf, len))
2687 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2688 rsp = cgi_error_memory();
2689 send_crunch_response(csp, rsp);
2694 /* Convert iob into something sed() can digest */
2695 if (JB_ERR_PARSE == get_server_headers(csp))
2699 /* Well, we tried our MS IIS/5
2700 * hack and it didn't work.
2701 * The header is incomplete
2702 * and there isn't anything
2703 * we can do about it.
2705 log_error(LOG_LEVEL_INFO,
2706 "MS IIS5 hack didn't produce valid headers.");
2711 /* Since we have to wait for
2712 * more from the server before
2713 * we can parse the headers
2714 * we just continue here.
2720 /* Did we actually get anything? */
2721 if (NULL == csp->headers->first)
2723 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2724 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2725 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2726 free_http_request(http);
2730 assert(csp->headers->first->str);
2732 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2733 strncmpic(csp->headers->first->str, "ICY", 3))
2736 * It doesn't look like a HTTP (or Shoutcast) response:
2737 * tell the client and log the problem.
2739 if (strlen(csp->headers->first->str) > 30)
2741 csp->headers->first->str[30] = '\0';
2743 log_error(LOG_LEVEL_ERROR,
2744 "Invalid server or forwarder response. Starts with: %s",
2745 csp->headers->first->str);
2746 log_error(LOG_LEVEL_CLF,
2747 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2748 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2749 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2750 free_http_request(http);
2755 * We have now received the entire server header,
2756 * filter it and send the result to the client
2758 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2760 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2762 hdr = list_to_text(csp->headers);
2765 /* FIXME Should handle error properly */
2766 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2769 if (crunch_response_triggered(csp, crunchers_light))
2772 * One of the tags created by a server-header
2773 * tagger triggered a crunch. We already
2774 * delivered the crunch response to the client
2775 * and are done here after cleaning up.
2780 /* Buffer and pcrs filter this if appropriate. */
2782 if (!http->ssl) /* We talk plaintext */
2784 content_filter = get_filter_function(csp);
2787 * Only write if we're not buffering for content modification
2789 if (!content_filter)
2791 /* write the server's (modified) header to
2792 * the client (along with anything else that
2793 * may be in the buffer)
2796 if (write_socket(csp->cfd, hdr, strlen(hdr))
2797 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2799 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2801 /* the write failed, so don't bother
2802 * mentioning it to the client...
2803 * it probably can't hear us anyway.
2809 byte_count += (size_t)len;
2812 /* we're finished with the server's header */
2817 /* If this was a MS IIS/5 hack then it means
2818 * the server has already closed the
2819 * connection. Nothing more to read. Time
2824 log_error(LOG_LEVEL_INFO,
2825 "Closed server connection detected with MS IIS5 hack enabled.");
2832 return; /* huh? we should never get here */
2835 if (csp->content_length == 0)
2838 * If Privoxy didn't recalculate the
2839 * Content-Lenght, byte_count is still
2842 csp->content_length = byte_count;
2845 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2846 csp->ip_addr_str, http->ocmd, csp->content_length);
2850 /*********************************************************************
2854 * Description : This is little more than chat. We only "serve" to
2855 * to close any socket that chat may have opened.
2858 * 1 : csp = Current client state (buffers, headers, etc...)
2862 *********************************************************************/
2864 void serve(struct client_state *csp)
2865 #else /* ifndef AMIGA */
2866 static void serve(struct client_state *csp)
2867 #endif /* def AMIGA */
2870 close_socket(csp->cfd);
2872 if (csp->sfd != JB_INVALID_SOCKET)
2874 close_socket(csp->sfd);
2877 csp->flags &= ~CSP_FLAG_ACTIVE;
2883 /*********************************************************************
2885 * Function : server_thread
2887 * Description : We only exist to call `serve' in a threaded environment.
2890 * 1 : data = Current client state (buffers, headers, etc...)
2892 * Returns : Always 0.
2894 *********************************************************************/
2895 static int32 server_thread(void *data)
2897 serve((struct client_state *) data);
2904 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2905 /*********************************************************************
2909 * Description : Print usage info & exit.
2911 * Parameters : Pointer to argv[0] for identifying ourselves
2915 *********************************************************************/
2916 static void usage(const char *myname)
2918 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2922 #endif /* defined(unix) */
2925 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2926 #endif /* defined(unix) */
2927 "[--version] [configfile]\n"
2928 "Aborting\n", myname);
2933 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2936 /*********************************************************************
2938 * Function : initialize_mutexes
2940 * Description : Prepares mutexes if mutex support is available.
2944 * Returns : Void, exits in case of errors.
2946 *********************************************************************/
2947 static void initialize_mutexes(void)
2951 #ifdef FEATURE_PTHREAD
2953 * Prepare global mutex semaphores
2955 err = pthread_mutex_init(&log_mutex, 0);
2957 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2960 * XXX: The assumptions below are a bit naive
2961 * and can cause locks that aren't necessary.
2963 * For example older FreeBSD versions (< 6.x?)
2964 * have no gethostbyname_r, but gethostbyname is
2967 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2968 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2969 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2971 * XXX: should we use a single mutex for
2972 * localtime() and gmtime() as well?
2974 #ifndef HAVE_GMTIME_R
2975 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2976 #endif /* ndef HAVE_GMTIME_R */
2978 #ifndef HAVE_LOCALTIME_R
2979 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2980 #endif /* ndef HAVE_GMTIME_R */
2983 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2984 #endif /* ndef HAVE_RANDOM */
2985 #endif /* FEATURE_PTHREAD */
2988 * TODO: mutex support for mingw32 would be swell.
2993 printf("Fatal error. Mutex initialization failed: %s.\n",
3002 /*********************************************************************
3006 * Description : Load the config file and start the listen loop.
3007 * This function is a lot more *sane* with the `load_config'
3008 * and `listen_loop' functions; although it stills does
3009 * a *little* too much for my taste.
3012 * 1 : argc = Number of parameters (including $0).
3013 * 2 : argv = Array of (char *)'s to the parameters.
3015 * Returns : 1 if : can't open config file, unrecognized directive,
3016 * stats requested in multi-thread mode, can't open the
3017 * log file, can't open the jar file, listen port is invalid,
3018 * any load fails, and can't bind port.
3020 * Else main never returns, the process must be signaled
3021 * to terminate execution. Or, on Windows, use the
3022 * "File", "Exit" menu option.
3024 *********************************************************************/
3026 int real_main(int argc, const char *argv[])
3028 int main(int argc, const char *argv[])
3032 unsigned int random_seed;
3034 struct passwd *pw = NULL;
3035 struct group *grp = NULL;
3038 char *pre_chroot_nslookup_to_load_resolver = NULL;
3045 #if !defined(_WIN32)
3053 * Parse the command line arguments
3055 * XXX: simply printing usage information in case of
3056 * invalid arguments isn't particularly user friendly.
3058 while (++argc_pos < argc)
3061 /* Check to see if the service must be installed or uninstalled */
3062 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3064 const char *pName = argv[argc_pos] + 9;
3067 exit( (install_service(pName)) ? 0 : 1 );
3069 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3071 const char *pName = argv[argc_pos] + 11;
3074 exit((uninstall_service(pName)) ? 0 : 1);
3076 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3078 bRunAsService = TRUE;
3079 w32_set_service_cwd();
3080 atexit(w32_service_exit_notify);
3083 #endif /* defined(_WIN32) */
3086 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3088 if (strcmp(argv[argc_pos], "--help") == 0)
3093 else if(strcmp(argv[argc_pos], "--version") == 0)
3095 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3101 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3106 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3108 if (++argc_pos == argc) usage(argv[0]);
3109 pidfile = strdup(argv[argc_pos]);
3112 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3114 if (++argc_pos == argc) usage(argv[argc_pos]);
3116 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3119 if (NULL == (grp = getgrnam(p)))
3121 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3125 if (NULL == (pw = getpwnam(argv[argc_pos])))
3127 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3130 if (p != NULL) *--p = '\0';
3133 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3135 if (++argc_pos == argc) usage(argv[0]);
3136 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3139 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3143 #endif /* defined(unix) */
3145 else if (argc_pos + 1 != argc)
3148 * This is neither the last command line
3149 * option, nor was it recognized before,
3150 * therefore it must be invalid.
3156 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3158 configfile = argv[argc_pos];
3161 } /* -END- while (more arguments) */
3164 if ( *configfile != '/' )
3166 char cwd[BUFFER_SIZE];
3168 size_t abs_file_size;
3170 /* make config-filename absolute here */
3171 if (NULL == getcwd(cwd, sizeof(cwd)))
3173 perror("failed to get current working directory");
3178 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3179 basedir = strdup(cwd);
3181 if (NULL == basedir ||
3182 NULL == (abs_file = malloc(abs_file_size)))
3184 perror("malloc failed");
3187 strlcpy(abs_file, basedir, abs_file_size);
3188 strlcat(abs_file, "/", abs_file_size );
3189 strlcat(abs_file, configfile, abs_file_size);
3190 configfile = abs_file;
3192 #endif /* defined unix */
3196 clients->next = NULL;
3198 /* XXX: factor out initialising after the next stable release. */
3201 #elif defined(_WIN32)
3205 /* Prepare mutexes if supported and necessary. */
3206 initialize_mutexes();
3208 /* Enable logging until further notice. */
3209 init_log_module(Argv[0]);
3211 random_seed = (unsigned int)time(NULL);
3213 srandom(random_seed);
3214 #elif defined (_WIN32)
3216 * See pick_from_range() in miscutil.c for details.
3218 log_error(LOG_LEVEL_INFO,
3219 "No thread-safe PRNG implemented for your platform. "
3220 "Using weak \'randomization\' factor which will "
3221 "limit the already questionable usefulness of "
3222 "header-time-randomizing actions (disabled by default).");
3225 #endif /* ifdef HAVE_RANDOM */
3228 * Unix signal handling
3230 * Catch the abort, interrupt and terminate signals for a graceful exit
3231 * Catch the hangup signal so the errlog can be reopened.
3232 * Ignore the broken pipe signals (FIXME: Why?)
3234 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3237 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3238 const int ignored_signals[] = { SIGPIPE, 0 };
3240 for (idx = 0; catched_signals[idx] != 0; idx++)
3242 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3243 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3245 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3246 #endif /* ifdef sun */
3248 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3252 for (idx = 0; ignored_signals[idx] != 0; idx++)
3254 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3256 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3261 #else /* ifdef _WIN32 */
3262 # ifdef _WIN_CONSOLE
3264 * We *are* in a windows console app.
3265 * Print a verbose messages about FAQ's and such
3267 printf("%s", win32_blurb);
3268 # endif /* def _WIN_CONSOLE */
3269 #endif /* def _WIN32 */
3272 /* Initialize the CGI subsystem */
3273 cgi_init_error_messages();
3276 * If runnig on unix and without the --nodaemon
3277 * option, become a daemon. I.e. fork, detach
3278 * from tty and get process group leadership
3291 if ( pid < 0 ) /* error */
3296 else if ( pid != 0 ) /* parent */
3301 * must check for errors
3302 * child died due to missing files aso
3305 wpid = waitpid( pid, &status, WNOHANG );
3314 /* Should be more portable, but not as well tested */
3319 #else /* ndef __FreeBSD__ */
3321 #endif /* ndef __FreeBSD__ */
3322 fd = open("/dev/tty", O_RDONLY);
3325 /* no error check here */
3326 ioctl( fd, TIOCNOTTY,0 );
3331 * stderr (fd 2) will be closed later on, when the
3332 * log file has been parsed.
3339 } /* -END- if (!no_daemon) */
3342 * As soon as we have written the PID file, we can switch
3343 * to the user and group ID indicated by the --user option
3349 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3351 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3357 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3359 /* Read the time zone file from /etc before doing chroot. */
3361 if (NULL != pre_chroot_nslookup_to_load_resolver
3362 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3364 /* Initialize resolver library. */
3365 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3367 if (chroot(pw->pw_dir) < 0)
3369 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3373 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3376 if (setuid(pw->pw_uid))
3378 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3382 char putenv_dummy[64];
3384 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3385 if (putenv(putenv_dummy) != 0)
3387 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3390 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3391 if (putenv(putenv_dummy) != 0)
3393 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3399 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3402 #endif /* defined unix */
3405 /* This will be FALSE unless the command line specified --service
3409 /* Yup, so now we must attempt to establish a connection
3410 * with the service dispatcher. This will only work if this
3411 * process was launched by the service control manager to
3412 * actually run as a service. If this isn't the case, i've
3413 * known it take around 30 seconds or so for the call to return.
3416 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3417 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3419 /* Service has run, and at this point is now being stopped, so just return */
3424 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3426 /* An error occurred. Usually it's because --service was wrongly specified
3427 * and we were unable to connect to the Service Control Dispatcher because
3428 * it wasn't expecting us and is therefore not listening.
3430 * For now, just continue below to call the listen_loop function.
3433 #endif /* def _WIN32 */
3443 /*********************************************************************
3445 * Function : bind_port_helper
3447 * Description : Bind the listen port. Handles logging, and aborts
3451 * 1 : config = Privoxy configuration. Specifies port
3454 * Returns : Port that was opened.
3456 *********************************************************************/
3457 static jb_socket bind_port_helper(struct configuration_spec * config)
3462 if (config->haddr == NULL)
3464 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3469 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3470 config->hport, config->haddr);
3473 result = bind_port(config->haddr, config->hport, &bfd);
3480 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3481 "There may be another Privoxy or some other "
3482 "proxy running on port %d",
3483 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3484 config->hport, config->hport);
3487 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3488 "The hostname is not resolvable",
3489 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3492 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3493 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3496 /* shouldn't get here */
3497 return JB_INVALID_SOCKET;
3500 config->need_bind = 0;
3507 /* Without this simple workaround we get this compiler warning from _beginthread
3508 * warning C4028: formal parameter 1 different from declaration
3510 void w32_service_listen_loop(void *p)
3514 #endif /* def _WIN32 */
3517 /*********************************************************************
3519 * Function : listen_loop
3521 * Description : bind the listen port and enter a "FOREVER" listening loop.
3527 *********************************************************************/
3528 static void listen_loop(void)
3530 struct client_state *csp = NULL;
3532 struct configuration_spec * config;
3534 config = load_config();
3536 bfd = bind_port_helper(config);
3538 #ifdef FEATURE_GRACEFUL_TERMINATION
3539 while (!g_terminate)
3544 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3545 while (waitpid(-1, NULL, WNOHANG) > 0)
3547 /* zombie children */
3549 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3552 * Free data that was used by died threads
3558 * Re-open the errlog after HUP signal
3560 if (received_hup_signal)
3562 init_error_log(Argv[0], config->logfile);
3563 received_hup_signal = 0;
3568 #ifdef FEATURE_COOKIE_JAR
3570 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3571 * already open. I think unload_configfile() is not being run, which should do
3572 * this work. Until that can get resolved, we'll use this workaround.
3576 if (csp->config->jar)
3578 fclose(csp->config->jar);
3579 csp->config->jar = NULL;
3581 #endif /* FEATURE_COOKIE_JAR */
3582 #endif /* __OS2__ */
3584 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3586 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3590 csp->flags |= CSP_FLAG_ACTIVE;
3591 csp->sfd = JB_INVALID_SOCKET;
3593 csp->config = config = load_config();
3595 if ( config->need_bind )
3598 * Since we were listening to the "old port", we will not see
3599 * a "listen" param change until the next IJB request. So, at
3600 * least 1 more request must be made for us to find the new
3601 * setting. I am simply closing the old socket and binding the
3604 * Which-ever is correct, we will serve 1 more page via the
3605 * old settings. This should probably be a "show-proxy-args"
3606 * request. This should not be a so common of an operation
3607 * that this will hurt people's feelings.
3612 bfd = bind_port_helper(config);
3615 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3617 if (!accept_connection(csp, bfd))
3619 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3632 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3635 #ifdef FEATURE_TOGGLE
3636 if (global_toggle_state)
3637 #endif /* def FEATURE_TOGGLE */
3639 csp->flags |= CSP_FLAG_TOGGLED_ON;
3642 if (run_loader(csp))
3644 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3645 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3649 if (block_acl(NULL,csp))
3651 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3652 close_socket(csp->cfd);
3656 #endif /* def FEATURE_ACL */
3658 /* add it to the list of clients */
3659 csp->next = clients->next;
3660 clients->next = csp;
3662 if (config->multi_threaded)
3666 /* this is a switch () statment in the C preprocessor - ugh */
3667 #undef SELECTED_ONE_OPTION
3669 /* Use Pthreads in preference to native code */
3670 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3671 #define SELECTED_ONE_OPTION
3673 pthread_t the_thread;
3674 pthread_attr_t attrs;
3676 pthread_attr_init(&attrs);
3677 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3678 errno = pthread_create(&the_thread, &attrs,
3679 (void * (*)(void *))serve, csp);
3680 child_id = errno ? -1 : 0;
3681 pthread_attr_destroy(&attrs);
3685 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3686 #define SELECTED_ONE_OPTION
3687 child_id = _beginthread(
3688 (void (*)(void *))serve,
3693 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3694 #define SELECTED_ONE_OPTION
3695 child_id = _beginthread(
3696 (void(* _Optlink)(void*))serve,
3702 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3703 #define SELECTED_ONE_OPTION
3705 thread_id tid = spawn_thread
3706 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3708 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3710 child_id = (int) tid;
3719 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3720 #define SELECTED_ONE_OPTION
3721 csp->cfd = ReleaseSocket(csp->cfd, -1);
3724 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3725 NP_Output, Output(),
3726 NP_CloseOutput, FALSE,
3727 NP_Name, (ULONG)"privoxy child",
3731 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3732 NP_Output, Output(),
3733 NP_CloseOutput, FALSE,
3734 NP_Name, (ULONG)"privoxy child",
3735 NP_StackSize, 200*1024,
3741 ((struct Task *)child_id)->tc_UserData = csp;
3742 Signal((struct Task *)child_id, SIGF_SINGLE);
3747 #if !defined(SELECTED_ONE_OPTION)
3750 /* This block is only needed when using fork().
3751 * When using threads, the server thread was
3752 * created and run by the call to _beginthread().
3754 if (child_id == 0) /* child */
3757 #ifdef FEATURE_TOGGLE
3758 int inherited_toggle_state = global_toggle_state;
3759 #endif /* def FEATURE_TOGGLE */
3764 * If we've been toggled or we've blocked the request, tell Mom
3767 #ifdef FEATURE_TOGGLE
3768 if (inherited_toggle_state != global_toggle_state)
3770 rc |= RC_FLAG_TOGGLED;
3772 #endif /* def FEATURE_TOGGLE */
3774 #ifdef FEATURE_STATISTICS
3775 if (csp->flags & CSP_FLAG_REJECTED)
3777 rc |= RC_FLAG_BLOCKED;
3779 #endif /* ndef FEATURE_STATISTICS */
3783 else if (child_id > 0) /* parent */
3785 /* in a fork()'d environment, the parent's
3786 * copy of the client socket and the CSP
3790 #if !defined(_WIN32) && !defined(__CYGWIN__)
3792 wait( &child_status );
3795 * Evaluate child's return code: If the child has
3796 * - been toggled, toggle ourselves
3797 * - blocked its request, bump up the stats counter
3800 #ifdef FEATURE_TOGGLE
3801 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3803 global_toggle_state = !global_toggle_state;
3805 #endif /* def FEATURE_TOGGLE */
3807 #ifdef FEATURE_STATISTICS
3809 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3813 #endif /* def FEATURE_STATISTICS */
3815 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3816 close_socket(csp->cfd);
3817 csp->flags &= ~CSP_FLAG_ACTIVE;
3821 #undef SELECTED_ONE_OPTION
3822 /* end of cpp switch () */
3824 if (child_id < 0) /* failed */
3826 char buf[BUFFER_SIZE];
3828 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3830 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3832 write_socket(csp->cfd, buf, strlen(buf));
3833 close_socket(csp->cfd);
3834 csp->flags &= ~CSP_FLAG_ACTIVE;
3845 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3847 /* Clean up. Aim: free all memory (no leaks) */
3848 #ifdef FEATURE_GRACEFUL_TERMINATION
3850 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3852 unload_current_config_file();
3853 unload_current_actions_file();
3854 unload_current_re_filterfile();
3855 #ifdef FEATURE_TRUST
3856 unload_current_trust_file();
3859 if (config->multi_threaded)
3866 } while ((clients->next != NULL) && (--i > 0));
3870 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3881 #ifdef FEATURE_COOKIE_JAR
3882 if (NULL != config->jar)
3884 fclose(config->jar);
3888 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3889 /* Cleanup - remove taskbar icon etc. */
3894 #endif /* FEATURE_GRACEFUL_TERMINATION */