1 const char jcc_rcs[] = "$Id: jcc.c,v 1.187 2008/09/07 12:35:05 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 1.187 2008/09/07 12:35:05 fabiankeil
37 * Add mutex lock support for _WIN32.
39 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
40 * Prepare for critical sections on Windows by adding a
41 * layer of indirection before the pthread mutex functions.
43 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
44 * Remove FEATURE_COOKIE_JAR.
46 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
47 * - Silence LLVM/Clang complaint.
48 * - Make received_hup_signal static.
49 * - Hide definitions for basedir, pidfile and received_hup_signal
50 * from __EMX__ as they only seem to be used in case of #ifdef unix.
52 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
53 * Accept Shoutcast responses again. Problem reported
54 * and fix suggested by Stefan in #2062860.
56 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
57 * Fix possible NULL-pointer dereference reported
58 * by din_a4 in #2003937. Pointy hat to me.
60 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
61 * Streamline sed()'s prototype and declare
62 * the header parse and add structures static.
64 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
65 * - Mark csp as immutable for send_crunch_response().
66 * - Fix comment spelling.
68 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
69 * Factor update_server_headers() out of sed(), ditch the
70 * first_run hack and make server_patterns_light static.
72 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
73 * Don't provide get_header() with the whole client state
74 * structure when it only needs access to csp->iob.
76 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
77 * Make the "read the rest of the headers" loop a bit more readable.
79 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
80 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
81 * - Remove useless comment.
83 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
84 * Fix comment grammar.
86 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
87 * Remove the pointless buffer in client_protocol_is_unsupported().
89 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
90 * Least-effort fix for bug #1821930 (reported by Lee):
91 * If the response doesn't look like HTTP,
92 * tell the client and log the problem.
94 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
95 * Don't pass the whole csp structure to flush_socket()
96 * when it only needs a file descriptor and a buffer.
98 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
99 * Remove kill-popups action.
101 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
102 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
104 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
105 * Remove the treat-forbidden-connects-like-blocks action. We now
106 * use the "blocked" page for forbidden CONNECT requests by default.
108 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
109 * Also use shiny new connect_port_is_forbidden() in jcc.c.
111 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
112 * Rename url_actions() to get_url_actions() and let it
113 * use the standard parameter ordering.
115 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
116 * Let forward_url() use the standard parameter ordering
117 * and mark its second parameter immutable.
119 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
120 * Remove the "Listening ... for local connections only" log message.
121 * Whether or not remote connections are able to reach Privoxy is up
122 * to the operating system.
124 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
125 * Prevent the log messages for CONNECT requests to unacceptable
126 * ports from printing the limit-connect argument as [null] if
127 * limit-connect hasn't been explicitly enabled.
129 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
130 * Make sure all console-mode apps get a usage() instance
132 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
133 * Reword NO_SERVER_DATA_RESPONSE to make it harder
134 * to misunderstand what the message is all about.
136 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
137 * Unbreak trustfile which previously didn't work without
138 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
140 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
141 * Plug memory leak. Spotted by Valgrind, triggered by
142 * Privoxy-Regression-Test feeding proxyfuzz.py.
144 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
145 * In the HTTP snipplets, refer to the client as client.
147 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
148 * Emit a log message when activating the MS IIS5 hack.
150 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
151 * Log the "weak randomization factor" warning only
152 * once for mingw32 and provide some more details.
154 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
155 * Initialize log module after initializing mutexes, future
156 * deadlocks in that code should now work cross-platform.
158 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
159 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
160 * Reported by Bert van Leeuwen in BR#1818808.
162 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
163 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
165 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
166 * When in daemon mode, close stderr after the configuration file has been
167 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
169 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
170 * - Fix a crash when parsing invalid requests whose first header
171 * is rejected by get_header(). Regression (re?)introduced
172 * in r1.143 by yours truly.
173 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
174 * client_cookie_adder() to make sure send-vanilla-wafer can be
175 * controlled through tags (and thus regression-tested).
177 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
178 * - Move get_filter_function() from jcc.c to filters.c
179 * so the filter functions can be static.
180 * - Don't bother filtering body-less responses.
182 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
183 * Execute content filters through execute_content_filter().
185 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
186 * Initialize req to NULL to make sure it's defined if the
187 * first read_socket() call fails. Reported by icmp30.
189 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
190 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
191 * extensive comments moved to user manual.
193 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
194 * Don't crash if a broken header filter wiped out the request line.
196 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
197 * Fix byte_count calculation in case of flushes
198 * and don't parse the server headers a second time.
200 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
201 * - If there's a connection problem after we already forwarded
202 * parts of the original content, just hang up. Fixes BR#1776724.
203 * - Fix warnings about unused code on mingw32.
204 * - In case of flushes, calculate the byte count
205 * less incorrectly (I think).
207 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
208 * Add some more prototypes for static functions.
210 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
211 * Comment out request_contains_null_bytes() until it's used again.
213 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
214 * #1763173 from Stefan Huehner: s@const static@static const@
215 * and declare some more functions static.
217 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
218 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
219 * and explain why they were rejected in the first place.
220 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
221 * CONNECT requests. The request line was missing.
222 * - Add two more XXX reminders as we don't have enough already.
224 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
225 * As Hal noticed, checking dispatch_cgi() as the last cruncher
226 * looks like a bug if CGI requests are blocked unintentionally,
227 * so don't do it unless the user enabled the new config option
228 * "allow-cgi-request-crunching".
230 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
231 * - Allow to rewrite the request destination behind the client's back.
232 * - Turn the weird-looking unconditional for loop that
233 * reads the client request into a conditional while loop.
234 * Move the stuff that only runs once out of the loop.
235 * - Move parts of chat(), server_content_type() and the
236 * necessary stuff to fix BR#1750917 into get_filter_function().
238 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
239 * Temporary workaround for BR#1730105.
241 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
242 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
243 * deadlocks and crashes on OpenBSD and possibly other OS with neither
244 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
245 * Thanks to Ralf Horstmann for report and solution.
247 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
248 * Add forward-override{} to change the forwarding settings through
249 * action sections. This is mainly interesting to forward different
250 * clients differently (for example based on User-Agent or request
253 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
254 * - Let usage() mention the --chroot parameter.
255 * - Use read_socket() consistently and always leave
256 * the last buffer byte alone, even in cases where
257 * null termination (currently) doesn't matter.
259 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
260 * - Fix config file loading on Unix if no config file is specified.
261 * Since r1.97 Privoxy would always interpret the last argument as
262 * config file, even if it's a valid command line option.
263 * - Abort in case of unrecognized command line options. Closes #1719696.
264 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
265 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
267 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
268 * - Don't rerun crunchers that only depend on the request URL.
269 * - Don't count redirects and CGI requests as "blocked requests".
271 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
272 * Support crunching based on tags created by server-header taggers.
274 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
275 * Make HTTP snippets static (again). Add a Content-Type for those
276 * with content so the browser doesn't guess it based on the URL.
278 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
279 * Move crunching and request line rebuilding out of chat().
281 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
282 * Introduce tags as alternative way to specify which
283 * actions apply to a request. At the moment tags can be
284 * created based on client and server headers.
286 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
287 * Don't CLF-log CONNECT requests twice.
289 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
290 * Log the source address for ACL-related connection drops.
292 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
293 * New config option: enforce-blocks.
295 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
296 * - Move null byte check into separate function.
297 * - Don't confuse the client with error pages
298 * if a CONNECT request was already confirmed.
300 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
301 * Speed up NULL byte escaping and only log the complete
302 * NULL byte requests with header debugging enabled.
304 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
305 * Answer requests that contain NULL bytes with
306 * a custom response instead of waiting for more
307 * data until the client eventually hangs up.
309 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
310 * - Move delivery and logging of crunched responses
311 * from chat() into send_crunch_response().
312 * - Display the reason for generating http_responses.
313 * - Log the content length for LOG_LEVEL_CLF correctly
314 * (still incorrect for some fixed responses).
315 * - Reword an incorrect comment about
316 * treat-forbidden-connects-like-blocks violating
318 * - Add some log messages.
320 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
321 * Move mutex initialization into separate
322 * function and exit in case of errors.
324 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
325 * - Start to reduce chat()'s line count and move
326 * parts of it into separate functions.
327 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
329 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
330 * - Add Proxy-Agent header to HTTP snippets that are
331 * supposed to reach HTTP clients only.
332 * - Made a few CONNECT log messages more descriptive.
333 * - Catch completely empty server responses (as seen
334 * with Tor's fake ".noconnect" top level domain).
335 * - Use shiny new "forwarding-failed" template for socks errors.
337 * Revision 1.118 2007/01/07 07:43:43 joergs
338 * AmigaOS4 support added.
340 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
341 * Added config option accept-intercepted-requests
342 * and disabled it by default.
344 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
345 * Reverted parts of my last commit
346 * to keep error handling working.
348 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
349 * Fixed gcc43 conversion warnings.
351 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
352 * Fix -pedantic ISO C warning about converting
353 * from function pointer to object pointer.
355 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
356 * Silence compiler warning I introduced with my last commit.
358 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
359 * Mutex protect rand() if POSIX threading
360 * is used, warn the user if that's not possible
361 * and stop using it on _WIN32 where it could
364 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
365 * Don't prevent core dumps by catching SIGABRT.
366 * It's rude and makes debugging unreasonable painful.
368 * Revision 1.110 2006/12/13 14:52:53 etresoft
369 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
371 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
372 * Privoxy is now able to run as intercepting
373 * proxy in combination with any packet filter
374 * that does the port redirection. The destination
375 * is extracted from the "Host:" header which
376 * should be available for nearly all requests.
378 * Moved HTTP snipplets into jcc.c.
379 * Added error message for gopher proxy requests.
381 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
382 * Only unlink the pidfile if it's actually used.
384 * Change order of interception checks to make
385 * it possible to block or redirect requests for
388 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
389 * Make pthread mutex locking more generic. Instead of
390 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
391 * and use mutex locking unless there is an _r function
392 * available. Better safe than sorry.
394 * Fixes "./configure --disable-pthread" and should result
395 * in less threading-related problems on pthread-using platforms,
396 * but it still doesn't fix BR#1122404.
398 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
399 * Move pthread.h inclusion from jcc.c to jcc.h.
400 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
402 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
403 * Don't exit after receiving the second SIGHUP on Solaris.
405 * Fixes BR 1052235, but the same problem may exist on other
406 * systems. Once 3.0.6 is out we should use sigset()
407 * where available and see if it breaks anything.
409 * Revision 1.104 2006/09/23 13:26:38 roro
410 * Replace TABs by spaces in source code.
412 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
413 * Fix +redirect{}. Didn't work with -fast-redirects.
415 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
416 * Respond with 400 and a short text message
417 * if the client tries to use Privoxy as FTP proxy.
419 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
420 * Make number of retries in case of forwarded-connect problems
421 * a config file option (forwarded-connect-retries) and use 0 as
424 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
425 * Set random(3) seed.
427 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
428 * Follow the OpenBSD port's lead and protect the resolve
429 * functions on OpenBSD as well.
431 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
432 * --user fix. Only use the user as group if no group is specified.
433 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
435 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
436 * Windows service (re-)integration
440 * --install[:service_name]
441 * --uninstall[:service_name]
444 * They work as follows:
445 * --install will create a service for you and then terminate.
446 * By default the service name will be "privoxy" (without the quotes).
447 * However you can run multiple services if you wish, just by adding
448 * a colon and then a name (no spaces).
450 * --uninstall follows the exact same rules a --install.
452 * --service is used when the program is executed by the service
453 * control manager, and in normal circumstances would never be
454 * used as a command line argument.
456 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
457 * Windows service integration
459 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
460 * Incorporate Fabian Keil's patch work:
461 http://www.fabiankeil.de/sourcecode/privoxy/
463 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
464 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
465 * with what was really the latest development (the v_3_0_branch branch)
467 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
468 * Thanks to Jindrich Makovicka for a race condition fix for the log
469 * file. The race condition remains for non-pthread implementations.
470 * Reference patch #1175720.
472 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
473 * Add the ability to check jpeg images for invalid
474 * lengths of comment blocks. Defensive strategy
475 * against the exploit:
476 * Microsoft Security Bulletin MS04-028
477 * Buffer Overrun in JPEG Processing (GDI+) Could
478 * Allow Code Execution (833987)
479 * Enabled with +inspect-jpegs in actions files.
481 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
482 * - Fixed usage info for non-unix platforms
483 * - Fixed small cmdline parsing bug
485 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
486 * Diagnostics: Now preserve the returncode of pthread_create
487 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
489 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
490 * No longer ignore SIGCHLD. Fixes bug #769381
492 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
493 * Close jarfile on graceful exit, remove stray line
495 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
496 * Cosmetics: Killed a warning, a typo and an allocation left at exit
498 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
499 * No longer rely on non-POSIX.1 extensions of getcwd().
502 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
503 * Replaced setenv() by posix-compliant putenv()
504 * Thanks to Neil McCalden (nmcc AT users.sf.net).
506 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
507 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
509 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
510 * Clean-up and extension of improvements for forked mode:
511 * - Child's return code now consists of flags RC_FLAG_*
512 * - Reporting toggle to parent now properly #ifdef'ed
513 * - Children now report blocking to parent. This enables
514 * statistics in forked mode
516 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
517 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
518 * Children now report having been toggled through _exit(17), parents
519 * watch for that code and toggle themselves if found.
521 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
522 * Wrapping all *_r functions (the non-_r versions of them) with
523 * mutex semaphores for OSX. Hopefully this will take care of all
524 * of those pesky crash reports.
526 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
527 * Fixed two mostly harmless mem leaks
529 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
530 * Fix: Head of global clients list now initialized to NULL
532 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
533 * Added basic support for OPTIONS and TRACE HTTP methods:
534 * - New interceptor direct_response() added in chat().
535 * - sed() moved to earlier in the process, so that the
536 * Host: header is evaluated before actions and forwarding
539 * Revision 1.92 2002/05/08 16:00:46 oes
540 * Chat's buffer handling:
541 * - Fixed bug with unchecked out-of-mem conditions
542 * while reading client request & server headers
543 * - No longer predict if the buffer limit will be exceeded
544 * in the next read -- check add_to_iob's new
545 * return code. If buffer couldn't be extended
546 * (policy or out-of-mem) while
547 * - reading from client: abort
548 * - reading server headers: send error page
549 * - buffering server body for filter: flush,
550 * and if that fails: send error page
552 * Revision 1.91 2002/04/08 20:35:58 swa
555 * Revision 1.90 2002/04/02 14:57:28 oes
556 * Made sending wafers independent of FEATURE_COOKIE_JAR
558 * Revision 1.89 2002/03/31 17:18:59 jongfoster
559 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
561 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
562 * More compiler warning message maintenance
564 * Revision 1.87 2002/03/26 22:29:54 swa
565 * we have a new homepage!
567 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
568 * Workaround for closing the jarfile before load_config() comes around again
570 * Revision 1.85 2002/03/24 15:23:33 jongfoster
573 * Revision 1.84 2002/03/24 13:25:43 swa
574 * name change related issues
576 * Revision 1.83 2002/03/16 23:54:06 jongfoster
577 * Adding graceful termination feature, to help look for memory leaks.
578 * If you enable this (which, by design, has to be done by hand
579 * editing config.h) and then go to http://i.j.b/die, then the program
580 * will exit cleanly after the *next* request. It should free all the
581 * memory that was used.
583 * Revision 1.82 2002/03/13 00:27:05 jongfoster
586 * Revision 1.81 2002/03/12 01:42:50 oes
587 * Introduced modular filters
589 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
590 * OS/2 port maintenance:
591 * - Fixed EMX build - it had decayed a little
592 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
593 * substituted a memset for now.
595 * Revision 1.79 2002/03/09 20:03:52 jongfoster
596 * - Making various functions return int rather than size_t.
597 * (Undoing a recent change). Since size_t is unsigned on
598 * Windows, functions like read_socket that return -1 on
599 * error cannot return a size_t.
601 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
602 * crashes, and also frequently caused JB to jump to 100%
603 * CPU and stay there. (Because it thought it had just
604 * read ((unsigned)-1) == 4Gb of data...)
606 * - The signature of write_socket has changed, it now simply
607 * returns success=0/failure=nonzero.
609 * - Trying to get rid of a few warnings --with-debug on
610 * Windows, I've introduced a new type "jb_socket". This is
611 * used for the socket file descriptors. On Windows, this
612 * is SOCKET (a typedef for unsigned). Everywhere else, it's
613 * an int. The error value can't be -1 any more, so it's
614 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
615 * Windows it maps to the #define INVALID_SOCKET.)
617 * - The signature of bind_port has changed.
619 * Revision 1.78 2002/03/08 21:35:04 oes
620 * Added optional group supplement to --user option. Will now use default group of user if no group given
622 * Revision 1.77 2002/03/07 03:52:06 oes
623 * - Fixed compiler warnings etc
624 * - Improved handling of failed DNS lookups
626 * Revision 1.76 2002/03/06 22:54:35 jongfoster
627 * Automated function-comment nitpicking.
629 * Revision 1.75 2002/03/06 10:02:19 oes
630 * Fixed stupid bug when --user was not given
632 * Revision 1.74 2002/03/06 00:49:31 jongfoster
633 * Fixing warning on Windows
634 * Making #ifdefs that refer to the same variable consistently
635 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
637 * Revision 1.73 2002/03/05 23:57:30 hal9
638 * Stray character 's' on line 1618 was breaking build.
640 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
641 * - Re-enable OS/2 building after new parms were added
642 * - Fix false out of memory report when resolving CGI templates when no IP
643 * address is available of failed attempt (a la no such domain)
645 * Revision 1.71 2002/03/05 18:13:56 oes
646 * Added --user option
648 * Revision 1.70 2002/03/05 04:52:42 oes
649 * Deleted non-errlog debugging code
651 * Revision 1.69 2002/03/04 23:50:00 jongfoster
652 * Splitting off bind_port() call into bind_port_helper(), with
655 * Revision 1.68 2002/03/04 20:17:32 oes
658 * Revision 1.67 2002/03/04 18:18:57 oes
659 * - Removed _DEBUG mode
660 * - Cleand up cmdline parsing
661 * - Introduced --no-daemon, --pidfile options
662 * - Cleaned up signal handling:
663 * - Terminate cleanly on INT, TERM and ABRT
664 * - Schedule logfile for re-opening on HUP
665 * - Ignore CHLD and PIPE
666 * - Leave the rest with their default handlers
667 * - Uniform handler registration
668 * - Added usage() function
669 * - Played styleguide police
671 * Revision 1.66 2002/03/03 15:06:55 oes
672 * Re-enabled automatic config reloading
674 * Revision 1.65 2002/03/03 14:49:11 oes
675 * Fixed CLF logging: Now uses client's original HTTP request
677 * Revision 1.64 2002/03/03 09:18:03 joergs
678 * Made jumbjuster work on AmigaOS again.
680 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
681 * Clean up a little CRLF unpleasantness that suddenly appeared
683 * Revision 1.62 2002/02/20 23:17:23 jongfoster
684 * Detecting some out-of memory conditions and exiting with a log message.
686 * Revision 1.61 2002/01/17 21:01:52 jongfoster
687 * Moving all our URL and URL pattern parsing code to urlmatch.c.
689 * Revision 1.60 2001/12/30 14:07:32 steudten
690 * - Add signal handling (unix)
691 * - Add SIGHUP handler (unix)
692 * - Add creation of pidfile (unix)
693 * - Add action 'top' in rc file (RH)
694 * - Add entry 'SIGNALS' to manpage
695 * - Add exit message to logfile (unix)
697 * Revision 1.59 2001/12/13 14:07:18 oes
698 * Fixed Bug: 503 error page now sent OK
700 * Revision 1.58 2001/11/30 23:37:24 jongfoster
701 * Renaming the Win32 config file to config.txt - this is almost the
702 * same as the corresponding UNIX name "config"
704 * Revision 1.57 2001/11/16 00:47:43 jongfoster
705 * Changing the tty-disconnection code to use setsid().
707 * Revision 1.56 2001/11/13 20:20:54 jongfoster
708 * Tabs->spaces, fixing a bug with missing {} around an if()
710 * Revision 1.55 2001/11/13 20:14:53 jongfoster
711 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
713 * Revision 1.54 2001/11/07 00:03:14 steudten
714 * Give reliable return value if an error
715 * occurs not just 0 with new daemon mode.
717 * Revision 1.53 2001/11/05 21:41:43 steudten
718 * Add changes to be a real daemon just for unix os.
719 * (change cwd to /, detach from controlling tty, set
720 * process group and session leader to the own process.
722 * Add some fatal-error log message for failed malloc().
723 * Add '-d' if compiled with 'configure --with-debug' to
724 * enable debug output.
726 * Revision 1.52 2001/10/26 20:11:20 jongfoster
727 * Fixing type mismatch
729 * Revision 1.51 2001/10/26 17:38:28 oes
732 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
733 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
734 * threads to call select() simultaneously. So, it's time to do a real, live,
735 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
736 * (native). Both versions will work, but using __OS2__ offers multi-threading.
738 * Revision 1.49 2001/10/23 21:41:35 jongfoster
739 * Added call to initialize the (statically-allocated of course)
740 * "out of memory" CGI response.
742 * Revision 1.48 2001/10/10 19:56:46 jongfoster
743 * Moving some code that wasn't cookie-related out of an #ifdef
746 * Revision 1.47 2001/10/10 16:44:36 oes
747 * Added CONNECT destination port limitation check
749 * Revision 1.46 2001/10/08 15:17:41 oes
750 * Re-enabled SSL forwarding
752 * Revision 1.45 2001/10/07 15:42:11 oes
753 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
755 * Moved downgrading of the HTTP version from parse_http_request to
756 * chat(), since we can't decide if it is necessary before we have
757 * determined the actions for the URL. The HTTP command is now
758 * *always* re-built so the repairs need no longer be special-cased.
760 * filter_popups now gets a csp pointer so it can raise the new
761 * CSP_FLAG_MODIFIED flag.
765 * Added configurable size limit for the IOB. If the IOB grows so
766 * large that the next read would exceed the limit, the header
767 * is generated, and the header & unfiltered buffer are flushed
768 * to the client. Chat then continues in non-buffering,
769 * non-filtering body mode.
771 * Revision 1.44 2001/10/02 18:13:57 oes
774 * Revision 1.43 2001/10/02 15:32:13 oes
775 * Moved generation of hdr
777 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
778 * Cleaning up 2 compiler warnings on OS/2.
780 * Revision 1.41 2001/09/16 17:05:14 jongfoster
781 * Removing unused #include showarg.h
783 * Revision 1.40 2001/09/16 15:41:45 jongfoster
784 * Fixing signed/unsigned comparison warning.
786 * Revision 1.39 2001/09/16 13:21:27 jongfoster
787 * Changes to use new list functions.
789 * Revision 1.38 2001/09/16 13:01:46 jongfoster
790 * Removing redundant function call that zeroed zalloc()'d memory.
792 * Revision 1.37 2001/09/10 11:12:24 oes
793 * Deleted unused variable
795 * Revision 1.36 2001/09/10 10:56:15 oes
796 * Silenced compiler warnings
798 * Revision 1.35 2001/07/31 14:44:22 oes
799 * Deleted unused size parameter from filter_popups()
801 * Revision 1.34 2001/07/30 22:08:36 jongfoster
802 * Tidying up #defines:
803 * - All feature #defines are now of the form FEATURE_xxx
804 * - Permanently turned off WIN_GUI_EDIT
805 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
807 * Revision 1.33 2001/07/29 19:32:00 jongfoster
808 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
810 * Revision 1.32 2001/07/29 18:47:05 jongfoster
811 * Adding missing #include "loadcfg.h"
813 * Revision 1.31 2001/07/29 12:17:48 oes
814 * Applied pthread fix by Paul Lieverse
816 * Revision 1.30 2001/07/25 22:57:13 jongfoster
817 * __BEOS__ no longer overrides FEATURE_PTHREAD.
818 * This is because FEATURE_PTHREAD will soon be widely used, so I
819 * want to keep it simple.
821 * Revision 1.29 2001/07/24 12:47:06 oes
822 * Applied BeOS support update by Eugenia
824 * Revision 1.28 2001/07/23 13:26:12 oes
825 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
827 * Revision 1.27 2001/07/19 19:09:47 haroon
828 * - Added code to take care of the situation where while processing the first
829 * server response (which includes the server header), after finding the end
830 * of the headers we were not looking past the end of the headers for
831 * content modification. I enabled it for filter_popups.
832 * Someone else should look to see if other similar operations should be
833 * done to the discarded portion of the buffer.
835 * Note 2001/07/20: No, the other content modification mechanisms will process
836 * the whole iob later anyway. --oes
838 * Revision 1.26 2001/07/18 12:31:36 oes
841 * Revision 1.25 2001/07/15 19:43:49 jongfoster
842 * Supports POSIX threads.
843 * Also removed some unused #includes.
845 * Revision 1.24 2001/07/13 14:00:40 oes
846 * - Generic content modification scheme:
847 * Each feature has its own applicability flag that is set
848 * from csp->action->flags.
849 * Replaced the "filtering" int flag , by a function pointer
850 * "content_filter" to the function that will do the content
851 * modification. If it is != NULL, the document will be buffered
852 * and processed through *content_filter, which must set
853 * csp->content_length and return a modified copy of the body
854 * or return NULL (on failiure).
855 * - Changed csp->is_text to the more generic bitmap csp->content_type
856 * which can currently take the valued CT_TEXT or CT_GIF
858 * - Removed all #ifdef PCRS
860 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
861 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
862 * the `block_acl' call. This prevents a compilation error when the user
863 * does not wish to use the "ACL" feature.
865 * Revision 1.22 2001/06/29 21:45:41 oes
866 * Indentation, CRLF->LF, Tab-> Space
868 * Revision 1.21 2001/06/29 13:29:36 oes
869 * - Cleaned up, improved comments
870 * - Unified all possible interceptors (CGI,
871 * block, trust, fast_redirect) in one
872 * place, with one (CGI) answer generation
873 * mechansim. Much clearer now.
874 * - Removed the GIF image generation, which
875 * is now done in filters.c:block_url()
876 * - Made error conditions like domain lookup
877 * failiure or (various) problems while talking
878 * to the server use cgi.c:error_response()
879 * instead of generating HTML/HTTP in chat() (yuck!)
880 * - Removed logentry from cancelled commit
882 * Revision 1.20 2001/06/09 10:55:28 jongfoster
883 * Changing BUFSIZ ==> BUFFER_SIZE
885 * Revision 1.19 2001/06/07 23:12:52 jongfoster
886 * Replacing function pointer in struct gateway with a directly
887 * called function forwarded_connect().
888 * Replacing struct gateway with struct forward_spec
890 * Revision 1.18 2001/06/03 19:12:16 oes
891 * introduced new cgi handling
893 * Revision 1.17 2001/06/01 20:07:23 jongfoster
894 * Now uses action +image-blocker{} rather than config->tinygif
896 * Revision 1.16 2001/06/01 18:49:17 jongfoster
897 * Replaced "list_share" with "list" - the tiny memory gain was not
898 * worth the extra complexity.
900 * Revision 1.15 2001/05/31 21:24:47 jongfoster
901 * Changed "permission" to "action" throughout.
902 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
903 * Moved vanilla wafer check into chat(), since we must now
904 * decide whether or not to add it based on the URL.
906 * Revision 1.14 2001/05/29 20:14:01 joergs
907 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
910 * Revision 1.13 2001/05/29 09:50:24 jongfoster
911 * Unified blocklist/imagelist/permissionslist.
912 * File format is still under discussion, but the internal changes
915 * Also modified interceptor behaviour:
916 * - We now intercept all URLs beginning with one of the following
917 * prefixes (and *only* these prefixes):
919 * * http://ijbswa.sf.net/config/
920 * * http://ijbswa.sourceforge.net/config/
921 * - New interceptors "home page" - go to http://i.j.b/ to see it.
922 * - Internal changes so that intercepted and fast redirect pages
923 * are not replaced with an image.
924 * - Interceptors now have the option to send a binary page direct
925 * to the client. (i.e. ijb-send-banner uses this)
926 * - Implemented show-url-info interceptor. (Which is why I needed
927 * the above interceptors changes - a typical URL is
928 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
929 * The previous mechanism would not have intercepted that, and
930 * if it had been intercepted then it then it would have replaced
933 * Revision 1.12 2001/05/27 22:17:04 oes
935 * - re_process_buffer no longer writes the modified buffer
936 * to the client, which was very ugly. It now returns the
937 * buffer, which it is then written by chat.
939 * - content_length now adjusts the Content-Length: header
940 * for modified documents rather than crunch()ing it.
941 * (Length info in csp->content_length, which is 0 for
942 * unmodified documents)
944 * - For this to work, sed() is called twice when filtering.
946 * Revision 1.11 2001/05/26 17:27:53 jongfoster
947 * Added support for CLF and fixed LOG_LEVEL_LOG.
948 * Also did CRLF->LF fix of my previous patch.
950 * Revision 1.10 2001/05/26 15:26:15 jongfoster
951 * ACL feature now provides more security by immediately dropping
952 * connections from untrusted hosts.
954 * Revision 1.9 2001/05/26 00:28:36 jongfoster
955 * Automatic reloading of config file.
956 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
957 * Most of the global variables have been moved to a new
958 * struct configuration_spec, accessed through csp->config->globalname
959 * Most of the globals remaining are used by the Win32 GUI.
961 * Revision 1.8 2001/05/25 22:43:18 jongfoster
962 * Fixing minor memory leak and buffer overflow.
964 * Revision 1.7 2001/05/25 22:34:30 jongfoster
967 * Revision 1.6 2001/05/23 00:13:58 joergs
968 * AmigaOS support fixed.
970 * Revision 1.5 2001/05/22 18:46:04 oes
972 * - Enabled filtering banners by size rather than URL
973 * by adding patterns that replace all standard banner
974 * sizes with the "Junkbuster" gif to the re_filterfile
976 * - Enabled filtering WebBugs by providing a pattern
977 * which kills all 1x1 images
979 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
980 * which is selected by the (nonstandard and therefore
981 * capital) letter 'U' in the option string.
982 * It causes the quantifiers to be ungreedy by default.
983 * Appending a ? turns back to greedy (!).
985 * - Added a new interceptor ijb-send-banner, which
986 * sends back the "Junkbuster" gif. Without imagelist or
987 * MSIE detection support, or if tinygif = 1, or the
988 * URL isn't recognized as an imageurl, a lame HTML
989 * explanation is sent instead.
991 * - Added new feature, which permits blocking remote
992 * script redirects and firing back a local redirect
994 * The feature is conditionally compiled, i.e. it
995 * can be disabled with --disable-fast-redirects,
996 * plus it must be activated by a "fast-redirects"
997 * line in the config file, has its own log level
998 * and of course wants to be displayed by show-proxy-args
999 * Note: Boy, all the #ifdefs in 1001 locations and
1000 * all the fumbling with configure.in and acconfig.h
1001 * were *way* more work than the feature itself :-(
1003 * - Because a generic redirect template was needed for
1004 * this, tinygif = 3 now uses the same.
1006 * - Moved GIFs, and other static HTTP response templates
1009 * - Some minor fixes
1011 * - Removed some >400 CRs again (Jon, you really worked
1014 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1015 * Made failure to bind() a fatal error.
1017 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1018 * Version 2.9.4 checkin.
1019 * - Merged popupfile and cookiefile, and added control over PCRS
1020 * filtering, in new "permissionsfile".
1021 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1022 * file error you now get a message box (in the Win32 GUI) rather
1023 * than the program exiting with no explanation.
1024 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1026 * - Removed tabs from "config"
1027 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1028 * - Bumped up version number.
1030 * Revision 1.2 2001/05/17 22:34:44 oes
1031 * - Added hint on GIF char array generation to jcc.c
1032 * - Cleaned CRLF's from the sources and related files
1033 * - Repaired logging for REF and FRC
1035 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1036 * Initial import of version 2.9.3 source tree
1039 *********************************************************************/
1045 #include <sys/types.h>
1054 # ifndef FEATURE_PTHREAD
1058 # include <windows.h>
1059 # include <process.h>
1060 # endif /* ndef FEATURE_PTHREAD */
1063 # ifndef _WIN_CONSOLE
1064 # include "w32log.h"
1065 # endif /* ndef _WIN_CONSOLE */
1066 # include "w32svrapi.h"
1068 #else /* ifndef _WIN32 */
1070 # if !defined (__OS2__)
1071 # include <unistd.h>
1072 # include <sys/wait.h>
1073 # endif /* ndef __OS2__ */
1074 # include <sys/time.h>
1075 # include <sys/stat.h>
1076 # include <sys/ioctl.h>
1079 #include <sys/termios.h>
1087 # include <signal.h>
1090 # include <socket.h> /* BeOS has select() for sockets only. */
1091 # include <OS.h> /* declarations for threads and stuff. */
1094 # if defined(__EMX__) || defined(__OS2__)
1095 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1100 #define bzero(B,N) memset(B,0x00,n)
1104 # include <select.h>
1109 #include "project.h"
1112 #include "filters.h"
1113 #include "loaders.h"
1114 #include "parsers.h"
1115 #include "miscutil.h"
1117 #include "jbsockets.h"
1118 #include "gateway.h"
1119 #include "actions.h"
1121 #include "loadcfg.h"
1122 #include "urlmatch.h"
1124 const char jcc_h_rcs[] = JCC_H_VERSION;
1125 const char project_h_rcs[] = PROJECT_H_VERSION;
1128 struct client_state clients[1];
1129 struct file_list files[1];
1131 #ifdef FEATURE_STATISTICS
1132 int urls_read = 0; /* total nr of urls read inc rejected */
1133 int urls_rejected = 0; /* total nr of urls rejected */
1134 #endif /* def FEATURE_STATISTICS */
1136 #ifdef FEATURE_GRACEFUL_TERMINATION
1137 int g_terminate = 0;
1140 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1141 static void sig_handler(int the_signal);
1143 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1144 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1145 static jb_err get_server_headers(struct client_state *csp);
1146 static const char *crunch_reason(const struct http_response *rsp);
1147 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1149 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1151 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1152 static jb_err change_request_destination(struct client_state *csp);
1153 static void chat(struct client_state *csp);
1154 static void serve(struct client_state *csp);
1155 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1156 static void usage(const char *myname);
1158 static void initialize_mutexes(void);
1159 static jb_socket bind_port_helper(struct configuration_spec *config);
1160 static void listen_loop(void);
1163 void serve(struct client_state *csp);
1164 #else /* ifndef AMIGA */
1165 static void serve(struct client_state *csp);
1166 #endif /* def AMIGA */
1169 static int32 server_thread(void *data);
1170 #endif /* def __BEOS__ */
1173 #define sleep(N) Sleep(((N) * 1000))
1177 #define sleep(N) DosSleep(((N) * 100))
1180 #ifdef MUTEX_LOCKS_AVAILABLE
1182 * XXX: Does the locking stuff really belong in this file?
1184 privoxy_mutex_t log_mutex;
1185 privoxy_mutex_t log_init_mutex;
1186 privoxy_mutex_t connection_reuse_mutex;
1188 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1189 privoxy_mutex_t resolver_mutex;
1190 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1192 #ifndef HAVE_GMTIME_R
1193 privoxy_mutex_t gmtime_mutex;
1194 #endif /* ndef HAVE_GMTIME_R */
1196 #ifndef HAVE_LOCALTIME_R
1197 privoxy_mutex_t localtime_mutex;
1198 #endif /* ndef HAVE_GMTIME_R */
1201 privoxy_mutex_t rand_mutex;
1202 #endif /* ndef HAVE_RANDOM */
1204 #endif /* def MUTEX_LOCKS_AVAILABLE */
1207 const char *basedir = NULL;
1208 const char *pidfile = NULL;
1209 static int received_hup_signal = 0;
1210 #endif /* defined unix */
1212 /* HTTP snipplets. */
1213 static const char CSUCCEED[] =
1214 "HTTP/1.0 200 Connection established\r\n"
1215 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1217 static const char CHEADER[] =
1218 "HTTP/1.0 400 Invalid header received from client\r\n"
1219 "Proxy-Agent: Privoxy " VERSION "\r\n"
1220 "Content-Type: text/plain\r\n"
1221 "Connection: close\r\n\r\n"
1222 "Invalid header received from client.\r\n";
1224 static const char FTP_RESPONSE[] =
1225 "HTTP/1.0 400 Invalid request received from client\r\n"
1226 "Content-Type: text/plain\r\n"
1227 "Connection: close\r\n\r\n"
1228 "Invalid request. Privoxy doesn't support FTP.\r\n";
1230 static const char GOPHER_RESPONSE[] =
1231 "HTTP/1.0 400 Invalid request received from client\r\n"
1232 "Content-Type: text/plain\r\n"
1233 "Connection: close\r\n\r\n"
1234 "Invalid request. Privoxy doesn't support gopher.\r\n";
1236 /* XXX: should be a template */
1237 static const char MISSING_DESTINATION_RESPONSE[] =
1238 "HTTP/1.0 400 Bad request received from client\r\n"
1239 "Proxy-Agent: Privoxy " VERSION "\r\n"
1240 "Content-Type: text/plain\r\n"
1241 "Connection: close\r\n\r\n"
1242 "Bad request. Privoxy was unable to extract the destination.\r\n";
1244 /* XXX: should be a template */
1245 static const char NO_SERVER_DATA_RESPONSE[] =
1246 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1247 "Proxy-Agent: Privoxy " VERSION "\r\n"
1248 "Content-Type: text/plain\r\n"
1249 "Connection: close\r\n\r\n"
1250 "Empty server or forwarder response.\r\n"
1251 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1253 /* XXX: should be a template */
1254 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1255 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1256 "Proxy-Agent: Privoxy " VERSION "\r\n"
1257 "Content-Type: text/plain\r\n"
1258 "Connection: close\r\n\r\n"
1259 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1262 /* XXX: should be a template */
1263 static const char NULL_BYTE_RESPONSE[] =
1264 "HTTP/1.0 400 Bad request received from client\r\n"
1265 "Proxy-Agent: Privoxy " VERSION "\r\n"
1266 "Content-Type: text/plain\r\n"
1267 "Connection: close\r\n\r\n"
1268 "Bad request. Null byte(s) before end of request.\r\n";
1271 /* XXX: should be a template */
1272 static const char MESSED_UP_REQUEST_RESPONSE[] =
1273 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1274 "Proxy-Agent: Privoxy " VERSION "\r\n"
1275 "Content-Type: text/plain\r\n"
1276 "Connection: close\r\n\r\n"
1277 "Bad request. Messed up with header filters.\r\n";
1279 /* A function to crunch a response */
1280 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1282 /* Crunch function flags */
1283 #define CF_NO_FLAGS 0
1284 /* Cruncher applies to forced requests as well */
1285 #define CF_IGNORE_FORCE 1
1286 /* Crunched requests are counted for the block statistics */
1287 #define CF_COUNT_AS_REJECT 2
1289 /* A crunch function and its flags */
1292 const crunch_func_ptr cruncher;
1296 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1298 /* Complete list of cruncher functions */
1299 static const struct cruncher crunchers_all[] = {
1300 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1301 { block_url, CF_COUNT_AS_REJECT },
1302 #ifdef FEATURE_TRUST
1303 { trust_url, CF_COUNT_AS_REJECT },
1304 #endif /* def FEATURE_TRUST */
1305 { redirect_url, CF_NO_FLAGS },
1306 { dispatch_cgi, CF_IGNORE_FORCE},
1310 /* Light version, used after tags are applied */
1311 static const struct cruncher crunchers_light[] = {
1312 { block_url, CF_COUNT_AS_REJECT },
1313 { redirect_url, CF_NO_FLAGS },
1319 * XXX: Don't we really mean
1325 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1326 /*********************************************************************
1328 * Function : sig_handler
1330 * Description : Signal handler for different signals.
1331 * Exit gracefully on TERM and INT
1332 * or set a flag that will cause the errlog
1333 * to be reopened by the main thread on HUP.
1336 * 1 : the_signal = the signal cause this function to call
1340 *********************************************************************/
1341 static void sig_handler(int the_signal)
1347 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1359 received_hup_signal = 1;
1365 * We shouldn't be here, unless we catch signals
1366 * in main() that we can't handle here!
1368 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1376 /*********************************************************************
1378 * Function : client_protocol_is_unsupported
1380 * Description : Checks if the client used a known unsupported
1381 * protocol and deals with it by sending an error
1385 * 1 : csp = Current client state (buffers, headers, etc...)
1386 * 2 : req = the first request line send by the client
1388 * Returns : TRUE if an error response has been generated, or
1389 * FALSE if the request doesn't look invalid.
1391 *********************************************************************/
1392 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1395 * If it's a FTP or gopher request, we don't support it.
1397 * These checks are better than nothing, but they might
1398 * not work in all configurations and some clients might
1399 * have problems digesting the answer.
1401 * They should, however, never cause more problems than
1402 * Privoxy's old behaviour (returning the misleading HTML
1405 * "Could not resolve http://(ftp|gopher)://example.org").
1407 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1409 const char *response = NULL;
1410 const char *protocol = NULL;
1412 if (!strncmpic(req, "GET ftp://", 10))
1414 response = FTP_RESPONSE;
1419 response = GOPHER_RESPONSE;
1420 protocol = "GOPHER";
1422 log_error(LOG_LEVEL_ERROR,
1423 "%s tried to use Privoxy as %s proxy: %s",
1424 csp->ip_addr_str, protocol, req);
1425 log_error(LOG_LEVEL_CLF,
1426 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1428 write_socket(csp->cfd, response, strlen(response));
1437 /*********************************************************************
1439 * Function : get_request_destination_elsewhere
1441 * Description : If the client's request was redirected into
1442 * Privoxy without the client's knowledge,
1443 * the request line lacks the destination host.
1445 * This function tries to get it elsewhere,
1446 * provided accept-intercepted-requests is enabled.
1448 * "Elsewhere" currently only means "Host: header",
1449 * but in the future we may ask the redirecting
1450 * packet filter to look the destination up.
1452 * If the destination stays unknown, an error
1453 * response is send to the client and headers
1454 * are freed so that chat() can return directly.
1457 * 1 : csp = Current client state (buffers, headers, etc...)
1458 * 2 : headers = a header list
1460 * Returns : JB_ERR_OK if the destination is now known, or
1461 * JB_ERR_PARSE if it isn't.
1463 *********************************************************************/
1464 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1468 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1470 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1471 " Privoxy isn't configured to accept intercepted requests.",
1472 csp->ip_addr_str, csp->http->cmd);
1473 /* XXX: Use correct size */
1474 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1475 csp->ip_addr_str, csp->http->cmd);
1477 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1478 destroy_list(headers);
1480 return JB_ERR_PARSE;
1482 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1484 /* Split the domain we just got for pattern matching */
1485 init_domain_components(csp->http);
1491 /* We can't work without destination. Go spread the news.*/
1493 req = list_to_text(headers);
1495 /* XXX: Use correct size */
1496 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1497 csp->ip_addr_str, csp->http->cmd);
1498 log_error(LOG_LEVEL_ERROR,
1499 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1500 csp->ip_addr_str, csp->http->cmd, req);
1503 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1504 destroy_list(headers);
1506 return JB_ERR_PARSE;
1509 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1510 * to get the destination IP address, use it as host directly
1511 * or do a reverse DNS lookup first.
1516 /*********************************************************************
1518 * Function : get_server_headers
1520 * Description : Parses server headers in iob and fills them
1521 * into csp->headers so that they can later be
1525 * 1 : csp = Current client state (buffers, headers, etc...)
1527 * Returns : JB_ERR_OK if everything went fine, or
1528 * JB_ERR_PARSE if the headers were incomplete.
1530 *********************************************************************/
1531 static jb_err get_server_headers(struct client_state *csp)
1533 int continue_hack_in_da_house = 0;
1536 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1541 * continue hack in da house. Ignore the ending of
1542 * this head and continue enlisting header lines.
1543 * The reason is described below.
1545 enlist(csp->headers, "");
1546 continue_hack_in_da_house = 0;
1549 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1552 * It's a bodyless continue response, don't
1553 * stop header parsing after reaching its end.
1555 * As a result Privoxy will concatenate the
1556 * next response's head and parse and deliver
1557 * the headers as if they belonged to one request.
1559 * The client will separate them because of the
1560 * empty line between them.
1562 * XXX: What we're doing here is clearly against
1563 * the intended purpose of the continue header,
1564 * and under some conditions (HTTP/1.0 client request)
1565 * it's a standard violation.
1567 * Anyway, "sort of against the spec" is preferable
1568 * to "always getting confused by Continue responses"
1569 * (Privoxy's behaviour before this hack was added)
1571 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1572 continue_hack_in_da_house = 1;
1574 else if (*header == '\0')
1577 * If the header is empty, but the Continue hack
1578 * isn't active, we can assume that we reached the
1579 * end of the buffer before we hit the end of the
1582 * Inform the caller an let it decide how to handle it.
1584 return JB_ERR_PARSE;
1587 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1590 * XXX: Should we quit the request and return a
1591 * out of memory error page instead?
1593 log_error(LOG_LEVEL_ERROR,
1594 "Out of memory while enlisting server headers. %s lost.",
1604 /*********************************************************************
1606 * Function : crunch_reason
1608 * Description : Translates the crunch reason code into a string.
1611 * 1 : rsp = a http_response
1613 * Returns : A string with the crunch reason or an error description.
1615 *********************************************************************/
1616 static const char *crunch_reason(const struct http_response *rsp)
1618 char * reason = NULL;
1620 assert(rsp != NULL);
1623 return "Internal error while searching for crunch reason";
1626 switch (rsp->reason)
1628 case RSP_REASON_UNSUPPORTED:
1629 reason = "Unsupported HTTP feature";
1631 case RSP_REASON_BLOCKED:
1634 case RSP_REASON_UNTRUSTED:
1635 reason = "Untrusted";
1637 case RSP_REASON_REDIRECTED:
1638 reason = "Redirected";
1640 case RSP_REASON_CGI_CALL:
1641 reason = "CGI Call";
1643 case RSP_REASON_NO_SUCH_DOMAIN:
1644 reason = "DNS failure";
1646 case RSP_REASON_FORWARDING_FAILED:
1647 reason = "Forwarding failed";
1649 case RSP_REASON_CONNECT_FAILED:
1650 reason = "Connection failure";
1652 case RSP_REASON_OUT_OF_MEMORY:
1653 reason = "Out of memory (may mask other reasons)";
1656 reason = "No reason recorded";
1664 /*********************************************************************
1666 * Function : send_crunch_response
1668 * Description : Delivers already prepared response for
1669 * intercepted requests, logs the interception
1670 * and frees the response.
1673 * 1 : csp = Current client state (buffers, headers, etc...)
1674 * 1 : rsp = Fully prepared response. Will be freed on exit.
1676 * Returns : Nothing.
1678 *********************************************************************/
1679 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1681 const struct http_request *http = csp->http;
1682 char status_code[4];
1684 assert(rsp != NULL);
1685 assert(rsp->head != NULL);
1690 * Not supposed to happen. If it does
1691 * anyway, treat it as an unknown error.
1693 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1694 /* return code doesn't matter */
1699 /* If rsp is still NULL, we have serious internal problems. */
1700 log_error(LOG_LEVEL_FATAL,
1701 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1705 * Extract the status code from the actual head
1706 * that was send to the client. It is the only
1707 * way to get it right for all requests, including
1708 * the fixed ones for out-of-memory problems.
1710 * A head starts like this: 'HTTP/1.1 200...'
1714 status_code[0] = rsp->head[9];
1715 status_code[1] = rsp->head[10];
1716 status_code[2] = rsp->head[11];
1717 status_code[3] = '\0';
1719 /* Write the answer to the client */
1720 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1721 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1723 /* There is nothing we can do about it. */
1724 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1727 /* Log that the request was crunched and why. */
1728 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1729 http->hostport, http->path, crunch_reason(rsp));
1730 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1731 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1733 /* Clean up and return */
1734 if (cgi_error_memory() != rsp)
1736 free_http_response(rsp);
1743 /*********************************************************************
1745 * Function : request_contains_null_bytes
1747 * Description : Checks for NULL bytes in the request and sends
1748 * an error message to the client if any were found.
1750 * XXX: currently not used, see comment in chat().
1753 * 1 : csp = Current client state (buffers, headers, etc...)
1754 * 2 : buf = Data from the client's request to check.
1755 * 3 : len = The data length.
1757 * Returns : TRUE if the request contained one or more NULL bytes, or
1760 *********************************************************************/
1761 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1763 size_t c_len; /* Request lenght when treated as C string */
1765 c_len = strlen(buf);
1770 * Null byte(s) found. Log the request,
1771 * return an error response and hang up.
1773 size_t tmp_len = c_len;
1778 * Replace NULL byte(s) with '°' characters
1779 * so the request can be logged as string.
1780 * XXX: Is there a better replacement character?
1783 tmp_len += strlen(buf+tmp_len);
1784 } while (tmp_len < len);
1786 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1787 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1788 log_error(LOG_LEVEL_HEADER,
1789 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1791 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1793 /* XXX: Log correct size */
1794 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1804 /*********************************************************************
1806 * Function : crunch_response_triggered
1808 * Description : Checks if the request has to be crunched,
1809 * and delivers the crunch response if necessary.
1812 * 1 : csp = Current client state (buffers, headers, etc...)
1813 * 2 : crunchers = list of cruncher functions to run
1815 * Returns : TRUE if the request was answered with a crunch response
1818 *********************************************************************/
1819 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1821 struct http_response *rsp = NULL;
1822 const struct cruncher *c;
1825 * If CGI request crunching is disabled,
1826 * check the CGI dispatcher out of order to
1827 * prevent unintentional blocks or redirects.
1829 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1830 && (NULL != (rsp = dispatch_cgi(csp))))
1832 /* Deliver, log and free the interception response. */
1833 send_crunch_response(csp, rsp);
1837 for (c = crunchers; c->cruncher != NULL; c++)
1840 * Check the cruncher if either Privoxy is toggled
1841 * on and the request isn't forced, or if the cruncher
1842 * applies to forced requests as well.
1844 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1845 !(csp->flags & CSP_FLAG_FORCED)) ||
1846 (c->flags & CF_IGNORE_FORCE))
1848 rsp = c->cruncher(csp);
1851 /* Deliver, log and free the interception response. */
1852 send_crunch_response(csp, rsp);
1853 #ifdef FEATURE_STATISTICS
1854 if (c->flags & CF_COUNT_AS_REJECT)
1856 csp->flags |= CSP_FLAG_REJECTED;
1858 #endif /* def FEATURE_STATISTICS */
1869 /*********************************************************************
1871 * Function : build_request_line
1873 * Description : Builds the HTTP request line.
1875 * If a HTTP forwarder is used it expects the whole URL,
1876 * web servers only get the path.
1879 * 1 : csp = Current client state (buffers, headers, etc...)
1880 * 2 : fwd = The forwarding spec used for the request
1881 * XXX: Should use http->fwd instead.
1882 * 3 : request_line = The old request line which will be replaced.
1884 * Returns : Nothing. Terminates in case of memory problems.
1886 *********************************************************************/
1887 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1889 struct http_request *http = csp->http;
1891 assert(http->ssl == 0);
1894 * Downgrade http version from 1.1 to 1.0
1895 * if +downgrade action applies.
1897 if ( (csp->action->flags & ACTION_DOWNGRADE)
1898 && (!strcmpic(http->ver, "HTTP/1.1")))
1901 http->ver = strdup("HTTP/1.0");
1903 if (http->ver == NULL)
1905 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1910 * Rebuild the request line.
1912 freez(*request_line);
1913 *request_line = strdup(http->gpc);
1914 string_append(request_line, " ");
1916 if (fwd->forward_host)
1918 string_append(request_line, http->url);
1922 string_append(request_line, http->path);
1924 string_append(request_line, " ");
1925 string_append(request_line, http->ver);
1927 if (*request_line == NULL)
1929 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1931 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1935 /*********************************************************************
1937 * Function : change_request_destination
1939 * Description : Parse a (rewritten) request line and regenerate
1940 * the http request data.
1943 * 1 : csp = Current client state (buffers, headers, etc...)
1945 * Returns : Forwards the parse_http_request() return code.
1946 * Terminates in case of memory problems.
1948 *********************************************************************/
1949 static jb_err change_request_destination(struct client_state *csp)
1951 struct http_request *http = csp->http;
1954 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1955 free_http_request(http);
1956 err = parse_http_request(csp->headers->first->str, http, csp);
1957 if (JB_ERR_OK != err)
1959 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1960 jb_err_to_string(err));
1964 /* XXX: ocmd is a misleading name */
1965 http->ocmd = strdup(http->cmd);
1966 if (http->ocmd == NULL)
1968 log_error(LOG_LEVEL_FATAL,
1969 "Out of memory copying rewritten HTTP request line");
1977 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
1978 /*********************************************************************
1980 * Function : server_response_is_complete
1982 * Description : Determines whether we should stop reading
1983 * from the server socket.
1986 * 1 : csp = Current client state (buffers, headers, etc...)
1988 * Returns : TRUE if the response is complete,
1991 *********************************************************************/
1992 static int server_response_is_complete(struct client_state *csp)
1994 int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
1996 if (!strcmpic(csp->http->gpc, "HEAD"))
1999 * "HEAD" implies no body, we are thus expecting
2000 * no content. XXX: incomplete "list" of methods?
2002 log_error(LOG_LEVEL_INFO, "Method %s implies no body.", csp->http->gpc);
2003 csp->expected_content_length = 0;
2004 content_length_known = TRUE;
2007 if (csp->http->status == 304)
2010 * Expect no body. XXX: incomplete "list" of status codes?
2012 log_error(LOG_LEVEL_INFO, "Status code %d implies no body.", csp->http->status);
2013 csp->expected_content_length = 0;
2014 content_length_known = TRUE;
2017 return (content_length_known && ((0 == csp->expected_content_length)
2018 || (csp->expected_content_length <= (csp->iob->eod - csp->iob->cur))));
2020 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2023 /*********************************************************************
2027 * Description : Once a connection to the client has been accepted,
2028 * this function is called (via serve()) to handle the
2029 * main business of the communication. When this
2030 * function returns, the caller must close the client
2033 * FIXME: chat is nearly thousand lines long.
2037 * 1 : csp = Current client state (buffers, headers, etc...)
2039 * Returns : Nothing.
2041 *********************************************************************/
2042 static void chat(struct client_state *csp)
2044 char buf[BUFFER_SIZE];
2052 int ms_iis5_hack = 0;
2053 size_t byte_count = 0;
2054 int forwarded_connect_retries = 0;
2055 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2056 const struct forward_spec * fwd;
2057 struct http_request *http;
2058 int len; /* for buffer sizes (and negative error codes) */
2061 /* Function that does the content filtering for the current request */
2062 filter_function_ptr content_filter = NULL;
2064 /* Skeleton for HTTP response, if we should intercept the request */
2065 struct http_response *rsp;
2067 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2068 struct list header_list;
2069 struct list *headers = &header_list;
2073 memset(buf, 0, sizeof(buf));
2076 * Read the client's request. Note that since we're not using select() we
2077 * could get blocked here if a client connected, then didn't say anything!
2082 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2084 if (len <= 0) break; /* error! */
2087 * If there is no memory left for buffering the
2088 * request, there is nothing we can do but hang up
2090 if (add_to_iob(csp, buf, len))
2095 req = get_header(csp->iob);
2097 } while ((NULL != req) && ('\0' == *req));
2099 if ((NULL != req) && ('\0' != *req))
2101 /* Request received. Validate and parse it. */
2105 * XXX: Temporary disabled to prevent problems
2106 * with POST requests whose bodies are allowed to
2107 * contain NULL bytes. BR#1730105.
2109 * The main purpose of this check is to properly
2110 * log stuff like BitTorrent traffic and other junk
2111 * that hits public proxies. It's not required for
2112 * Privoxy to functions as those requests are discarded
2115 * It probably should be rewritten to only check
2116 * the head of the request. Another option would
2117 * be to let all POST requests pass, although that
2118 * may not be good enough.
2120 if (request_contains_null_bytes(csp, buf, len))
2122 /* NULL bytes found and dealt with, just hang up. */
2127 /* Does the request line look invalid? */
2128 if (client_protocol_is_unsupported(csp, req))
2131 * Yes. The request has already been
2132 * answered with a error response, the buffers
2133 * were freed and we're done with chatting.
2138 #ifdef FEATURE_FORCE_LOAD
2140 * If this request contains the FORCE_PREFIX and blocks
2141 * aren't enforced, get rid of it and set the force flag.
2143 if (strstr(req, FORCE_PREFIX))
2145 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2147 log_error(LOG_LEVEL_FORCE,
2148 "Ignored force prefix in request: \"%s\".", req);
2152 strclean(req, FORCE_PREFIX);
2153 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2154 csp->flags |= CSP_FLAG_FORCED;
2158 #endif /* def FEATURE_FORCE_LOAD */
2159 err = parse_http_request(req, http, csp);
2160 if (JB_ERR_OK != err)
2162 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2168 if (http->cmd == NULL)
2170 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2171 /* XXX: Use correct size */
2172 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2173 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2175 free_http_request(http);
2179 /* grab the rest of the client's headers */
2183 p = get_header(csp->iob);
2187 /* There are no additional headers to read. */
2194 * We didn't receive a complete header
2195 * line yet, get the rest of it.
2197 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2200 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2201 destroy_list(headers);
2205 if (add_to_iob(csp, buf, len))
2208 * If there is no memory left for buffering the
2209 * request, there is nothing we can do but hang up
2211 destroy_list(headers);
2218 * We were able to read a complete
2219 * header and can finaly enlist it.
2226 if (http->host == NULL)
2229 * If we still don't know the request destination,
2230 * the request is invalid or the client uses
2231 * Privoxy without its knowledge.
2233 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2236 * Our attempts to get the request destination
2237 * elsewhere failed or Privoxy is configured
2238 * to only accept proxy requests.
2240 * An error response has already been send
2241 * and we're done here.
2248 * Determine the actions for this URL
2250 #ifdef FEATURE_TOGGLE
2251 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2253 /* Most compatible set of actions (i.e. none) */
2254 init_current_action(csp->action);
2257 #endif /* ndef FEATURE_TOGGLE */
2259 get_url_actions(csp, http);
2263 * Save a copy of the original request for logging
2265 http->ocmd = strdup(http->cmd);
2267 if (http->ocmd == NULL)
2269 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2272 enlist(csp->headers, http->cmd);
2274 /* Append the previously read headers */
2275 list_append_list_unique(csp->headers, headers);
2276 destroy_list(headers);
2278 err = sed(csp, FILTER_CLIENT_HEADERS);
2279 if (JB_ERR_OK != err)
2281 assert(err == JB_ERR_PARSE);
2282 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2284 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2286 /* Check request line for rewrites. */
2287 if ((NULL == csp->headers->first->str)
2288 || (strcmp(http->cmd, csp->headers->first->str) &&
2289 (JB_ERR_OK != change_request_destination(csp))))
2292 * A header filter broke the request line - bail out.
2294 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2295 /* XXX: Use correct size */
2296 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2297 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2299 free_http_request(http);
2303 /* decide how to route the HTTP request */
2304 fwd = forward_url(csp, http);
2307 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2308 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2312 * build the http request to send to the server
2313 * we have to do one of the following:
2315 * create = use the original HTTP request to create a new
2316 * HTTP request that has either the path component
2317 * without the http://domainspec (w/path) or the
2318 * full orininal URL (w/url)
2319 * Note that the path and/or the HTTP version may
2320 * have been altered by now.
2322 * connect = Open a socket to the host:port of the server
2323 * and short-circuit server and client socket.
2325 * pass = Pass the request unchanged if forwarding a CONNECT
2326 * request to a parent proxy. Note that we'll be sending
2327 * the CFAIL message ourselves if connecting to the parent
2328 * fails, but we won't send a CSUCCEED message if it works,
2329 * since that would result in a double message (ours and the
2330 * parent's). After sending the request to the parent, we simply
2333 * here's the matrix:
2336 * +--------+--------+
2338 * 0 | create | connect|
2340 * Forwarding +--------+--------+
2342 * 1 | create | pass |
2344 * +--------+--------+
2348 if (http->ssl && connect_port_is_forbidden(csp))
2350 const char *acceptable_connect_ports =
2351 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2352 assert(NULL != acceptable_connect_ports);
2353 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2354 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2355 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2356 csp->action->flags |= ACTION_BLOCK;
2362 freez(csp->headers->first->str);
2363 build_request_line(csp, fwd, &csp->headers->first->str);
2366 hdr = list_to_text(csp->headers);
2369 /* FIXME Should handle error properly */
2370 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2374 * We have a request. Check if one of the crunchers wants it.
2376 if (crunch_response_triggered(csp, crunchers_all))
2379 * Yes. The client got the crunch response
2380 * and we are done here after cleaning up.
2383 list_remove_all(csp->headers);
2389 * The headers can't be removed earlier because
2390 * they were still needed for the referrer check
2391 * in case of CGI crunches.
2393 * XXX: Would it be worth to move the referrer check
2394 * into client_referrer() and set a flag if it's trusted?
2396 list_remove_all(csp->headers);
2398 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2400 if (fwd->forward_host)
2402 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2403 fwd->forward_host, fwd->forward_port, http->hostport);
2407 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2410 /* here we connect to the server, gateway, or the forwarder */
2412 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2413 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2415 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2416 forwarded_connect_retries, http->hostport);
2419 if (csp->sfd == JB_INVALID_SOCKET)
2421 if (fwd->type != SOCKS_NONE)
2424 rsp = error_response(csp, "forwarding-failed", errno);
2426 else if (errno == EINVAL)
2428 rsp = error_response(csp, "no-such-domain", errno);
2432 rsp = error_response(csp, "connect-failed", errno);
2433 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2438 /* Write the answer to the client */
2441 send_crunch_response(csp, rsp);
2448 if (fwd->forward_host || (http->ssl == 0))
2450 /* write the client's (modified) header to the server
2451 * (along with anything else that may be in the buffer)
2454 if (write_socket(csp->sfd, hdr, strlen(hdr))
2455 || (flush_socket(csp->sfd, csp->iob) < 0))
2457 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2460 rsp = error_response(csp, "connect-failed", errno);
2464 send_crunch_response(csp, rsp);
2474 * We're running an SSL tunnel and we're not forwarding,
2475 * so just send the "connect succeeded" message to the
2476 * client, flush the rest, and get out of the way.
2478 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2486 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2488 /* we're finished with the client's header */
2491 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2493 /* pass data between the client and server
2494 * until one or the other shuts down the connection.
2503 * FD_ZERO here seems to point to an errant macro which crashes.
2504 * So do this by hand for now...
2506 memset(&rfds,0x00,sizeof(fd_set));
2510 FD_SET(csp->cfd, &rfds);
2511 FD_SET(csp->sfd, &rfds);
2513 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2514 if (server_body && server_response_is_complete(csp))
2516 log_error(LOG_LEVEL_CONNECT,
2517 "Stopped reading from server. Expected content length: %d. "
2518 "Actual content length: %d. Most recently received: %d.",
2519 csp->expected_content_length, (csp->iob->eod - csp->iob->cur), len);
2522 * XXX: should not jump around,
2523 * chat() is complicated enough already.
2527 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2529 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2533 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2537 /* this is the body of the browser's request
2538 * just read it and write it.
2541 if (FD_ISSET(csp->cfd, &rfds))
2543 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2547 break; /* "game over, man" */
2550 if (write_socket(csp->sfd, buf, (size_t)len))
2552 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2559 * The server wants to talk. It could be the header or the body.
2560 * If `hdr' is null, then it's the header otherwise it's the body.
2561 * FIXME: Does `hdr' really mean `host'? No.
2565 if (FD_ISSET(csp->sfd, &rfds))
2568 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2572 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2574 if (http->ssl && (fwd->forward_host == NULL))
2577 * Just hang up. We already confirmed the client's CONNECT
2578 * request with status code 200 and unencrypted content is
2579 * no longer welcome.
2581 log_error(LOG_LEVEL_ERROR,
2582 "CONNECT already confirmed. Unable to tell the client about the problem.");
2585 else if (byte_count)
2588 * Just hang up. We already transmitted the original headers
2589 * and parts of the original content and therefore missed the
2590 * chance to send an error message (without risking data corruption).
2592 * XXX: we could retry with a fancy range request here.
2594 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2595 "Unable to tell the client about the problem.");
2599 rsp = error_response(csp, "connect-failed", errno);
2603 send_crunch_response(csp, rsp);
2609 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2611 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2613 /* Add a trailing zero. This lets filter_popups
2614 * use string operations.
2618 /* Normally, this would indicate that we've read
2619 * as much as the server has sent us and we can
2620 * close the client connection. However, Microsoft
2621 * in its wisdom has released IIS/5 with a bug that
2622 * prevents it from sending the trailing \r\n in
2623 * a 302 redirect header (and possibly other headers).
2624 * To work around this if we've haven't parsed
2625 * a full header we'll append a trailing \r\n
2626 * and see if this now generates a valid one.
2628 * This hack shouldn't have any impacts. If we've
2629 * already transmitted the header or if this is a
2630 * SSL connection, then we won't bother with this
2631 * hack. So we only work on partially received
2632 * headers. If we append a \r\n and this still
2633 * doesn't generate a valid header, then we won't
2634 * transmit anything to the client.
2639 if (server_body || http->ssl)
2642 * If we have been buffering up the document,
2643 * now is the time to apply content modification
2644 * and send the result to the client.
2648 p = execute_content_filter(csp, content_filter);
2650 * If the content filter fails, use the original
2651 * buffer and length.
2652 * (see p != NULL ? p : csp->iob->cur below)
2656 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2659 if (JB_ERR_OK != update_server_headers(csp))
2661 log_error(LOG_LEVEL_FATAL,
2662 "Failed to update server headers. after filtering.");
2665 hdr = list_to_text(csp->headers);
2668 /* FIXME Should handle error properly */
2669 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2672 if (write_socket(csp->cfd, hdr, strlen(hdr))
2673 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2675 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2685 break; /* "game over, man" */
2689 * This is NOT the body, so
2690 * Let's pretend the server just sent us a blank line.
2692 snprintf(buf, sizeof(buf), "\r\n");
2693 len = (int)strlen(buf);
2696 * Now, let the normal header parsing algorithm below do its
2697 * job. If it fails, we'll exit instead of continuing.
2704 * If this is an SSL connection or we're in the body
2705 * of the server document, just write it to the client,
2706 * unless we need to buffer the body for later content-filtering
2709 if (server_body || http->ssl)
2714 * If there is no memory left for buffering the content, or the buffer limit
2715 * has been reached, switch to non-filtering mode, i.e. make & write the
2716 * header, flush the iob and buf, and get out of the way.
2718 if (add_to_iob(csp, buf, len))
2723 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2725 hdr = list_to_text(csp->headers);
2729 * Memory is too tight to even generate the header.
2730 * Send our static "Out-of-memory" page.
2732 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2733 rsp = cgi_error_memory();
2734 send_crunch_response(csp, rsp);
2738 hdrlen = strlen(hdr);
2740 if (write_socket(csp->cfd, hdr, hdrlen)
2741 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2742 || (write_socket(csp->cfd, buf, (size_t)len)))
2744 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2751 * Reset the byte_count to the amount of bytes
2752 * we just flushed. len will be added a few lines below,
2753 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2755 byte_count = (size_t)flushed;
2757 content_filter = NULL;
2763 if (write_socket(csp->cfd, buf, (size_t)len))
2765 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2769 byte_count += (size_t)len;
2774 /* we're still looking for the end of the
2775 * server's header ... (does that make header
2776 * parsing an "out of body experience" ?
2780 * buffer up the data we just read. If that fails,
2781 * there's little we can do but send our static
2782 * out-of-memory page.
2784 if (add_to_iob(csp, buf, len))
2786 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2787 rsp = cgi_error_memory();
2788 send_crunch_response(csp, rsp);
2793 /* Convert iob into something sed() can digest */
2794 if (JB_ERR_PARSE == get_server_headers(csp))
2798 /* Well, we tried our MS IIS/5
2799 * hack and it didn't work.
2800 * The header is incomplete
2801 * and there isn't anything
2802 * we can do about it.
2804 log_error(LOG_LEVEL_INFO,
2805 "MS IIS5 hack didn't produce valid headers.");
2810 /* Since we have to wait for
2811 * more from the server before
2812 * we can parse the headers
2813 * we just continue here.
2819 /* Did we actually get anything? */
2820 if (NULL == csp->headers->first)
2822 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2823 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2824 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2825 free_http_request(http);
2829 assert(csp->headers->first->str);
2831 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2832 strncmpic(csp->headers->first->str, "ICY", 3))
2835 * It doesn't look like a HTTP (or Shoutcast) response:
2836 * tell the client and log the problem.
2838 if (strlen(csp->headers->first->str) > 30)
2840 csp->headers->first->str[30] = '\0';
2842 log_error(LOG_LEVEL_ERROR,
2843 "Invalid server or forwarder response. Starts with: %s",
2844 csp->headers->first->str);
2845 log_error(LOG_LEVEL_CLF,
2846 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2847 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2848 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2849 free_http_request(http);
2854 * We have now received the entire server header,
2855 * filter it and send the result to the client
2857 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2859 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2861 hdr = list_to_text(csp->headers);
2864 /* FIXME Should handle error properly */
2865 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2868 if (crunch_response_triggered(csp, crunchers_light))
2871 * One of the tags created by a server-header
2872 * tagger triggered a crunch. We already
2873 * delivered the crunch response to the client
2874 * and are done here after cleaning up.
2879 /* Buffer and pcrs filter this if appropriate. */
2881 if (!http->ssl) /* We talk plaintext */
2883 content_filter = get_filter_function(csp);
2886 * Only write if we're not buffering for content modification
2888 if (!content_filter)
2890 /* write the server's (modified) header to
2891 * the client (along with anything else that
2892 * may be in the buffer)
2895 if (write_socket(csp->cfd, hdr, strlen(hdr))
2896 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2898 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2900 /* the write failed, so don't bother
2901 * mentioning it to the client...
2902 * it probably can't hear us anyway.
2908 byte_count += (size_t)len;
2911 /* we're finished with the server's header */
2916 /* If this was a MS IIS/5 hack then it means
2917 * the server has already closed the
2918 * connection. Nothing more to read. Time
2923 log_error(LOG_LEVEL_INFO,
2924 "Closed server connection detected with MS IIS5 hack enabled.");
2931 return; /* huh? we should never get here */
2934 if (csp->content_length == 0)
2937 * If Privoxy didn't recalculate the
2938 * Content-Lenght, byte_count is still
2941 csp->content_length = byte_count;
2944 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2945 csp->ip_addr_str, http->ocmd, csp->content_length);
2949 /*********************************************************************
2953 * Description : This is little more than chat. We only "serve" to
2954 * to close any socket that chat may have opened.
2957 * 1 : csp = Current client state (buffers, headers, etc...)
2961 *********************************************************************/
2963 void serve(struct client_state *csp)
2964 #else /* ifndef AMIGA */
2965 static void serve(struct client_state *csp)
2966 #endif /* def AMIGA */
2969 close_socket(csp->cfd);
2971 if (csp->sfd != JB_INVALID_SOCKET)
2973 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2974 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2976 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
2980 forget_connection(csp->sfd);
2981 close_socket(csp->sfd);
2984 close_socket(csp->sfd);
2985 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
2988 csp->flags &= ~CSP_FLAG_ACTIVE;
2994 /*********************************************************************
2996 * Function : server_thread
2998 * Description : We only exist to call `serve' in a threaded environment.
3001 * 1 : data = Current client state (buffers, headers, etc...)
3003 * Returns : Always 0.
3005 *********************************************************************/
3006 static int32 server_thread(void *data)
3008 serve((struct client_state *) data);
3015 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3016 /*********************************************************************
3020 * Description : Print usage info & exit.
3022 * Parameters : Pointer to argv[0] for identifying ourselves
3026 *********************************************************************/
3027 static void usage(const char *myname)
3029 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3033 #endif /* defined(unix) */
3036 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3037 #endif /* defined(unix) */
3038 "[--version] [configfile]\n"
3039 "Aborting\n", myname);
3044 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3047 #ifdef MUTEX_LOCKS_AVAILABLE
3048 /*********************************************************************
3050 * Function : privoxy_mutex_lock
3052 * Description : Locks a mutex.
3055 * 1 : mutex = The mutex to lock.
3057 * Returns : Void. May exit in case of errors.
3059 *********************************************************************/
3060 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3062 #ifdef FEATURE_PTHREAD
3063 int err = pthread_mutex_lock(mutex);
3066 if (mutex != &log_mutex)
3068 log_error(LOG_LEVEL_FATAL,
3069 "Mutex locking failed: %s.\n", strerror(err));
3074 EnterCriticalSection(mutex);
3075 #endif /* def FEATURE_PTHREAD */
3079 /*********************************************************************
3081 * Function : privoxy_mutex_unlock
3083 * Description : Unlocks a mutex.
3086 * 1 : mutex = The mutex to unlock.
3088 * Returns : Void. May exit in case of errors.
3090 *********************************************************************/
3091 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3093 #ifdef FEATURE_PTHREAD
3094 int err = pthread_mutex_unlock(mutex);
3097 if (mutex != &log_mutex)
3099 log_error(LOG_LEVEL_FATAL,
3100 "Mutex unlocking failed: %s.\n", strerror(err));
3105 LeaveCriticalSection(mutex);
3106 #endif /* def FEATURE_PTHREAD */
3110 /*********************************************************************
3112 * Function : privoxy_mutex_init
3114 * Description : Prepares a mutex.
3117 * 1 : mutex = The mutex to initialize.
3119 * Returns : Void. May exit in case of errors.
3121 *********************************************************************/
3122 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3124 #ifdef FEATURE_PTHREAD
3125 int err = pthread_mutex_init(mutex, 0);
3128 printf("Fatal error. Mutex initialization failed: %s.\n",
3133 InitializeCriticalSection(mutex);
3134 #endif /* def FEATURE_PTHREAD */
3136 #endif /* def MUTEX_LOCKS_AVAILABLE */
3138 /*********************************************************************
3140 * Function : initialize_mutexes
3142 * Description : Prepares mutexes if mutex support is available.
3146 * Returns : Void, exits in case of errors.
3148 *********************************************************************/
3149 static void initialize_mutexes(void)
3151 #ifdef MUTEX_LOCKS_AVAILABLE
3153 * Prepare global mutex semaphores
3155 privoxy_mutex_init(&log_mutex);
3156 privoxy_mutex_init(&log_init_mutex);
3157 privoxy_mutex_init(&connection_reuse_mutex);
3160 * XXX: The assumptions below are a bit naive
3161 * and can cause locks that aren't necessary.
3163 * For example older FreeBSD versions (< 6.x?)
3164 * have no gethostbyname_r, but gethostbyname is
3167 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3168 privoxy_mutex_init(&resolver_mutex);
3169 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3171 * XXX: should we use a single mutex for
3172 * localtime() and gmtime() as well?
3174 #ifndef HAVE_GMTIME_R
3175 privoxy_mutex_init(&gmtime_mutex);
3176 #endif /* ndef HAVE_GMTIME_R */
3178 #ifndef HAVE_LOCALTIME_R
3179 privoxy_mutex_init(&localtime_mutex);
3180 #endif /* ndef HAVE_GMTIME_R */
3183 privoxy_mutex_init(&rand_mutex);
3184 #endif /* ndef HAVE_RANDOM */
3185 #endif /* def MUTEX_LOCKS_AVAILABLE */
3189 /*********************************************************************
3193 * Description : Load the config file and start the listen loop.
3194 * This function is a lot more *sane* with the `load_config'
3195 * and `listen_loop' functions; although it stills does
3196 * a *little* too much for my taste.
3199 * 1 : argc = Number of parameters (including $0).
3200 * 2 : argv = Array of (char *)'s to the parameters.
3202 * Returns : 1 if : can't open config file, unrecognized directive,
3203 * stats requested in multi-thread mode, can't open the
3204 * log file, can't open the jar file, listen port is invalid,
3205 * any load fails, and can't bind port.
3207 * Else main never returns, the process must be signaled
3208 * to terminate execution. Or, on Windows, use the
3209 * "File", "Exit" menu option.
3211 *********************************************************************/
3213 int real_main(int argc, const char *argv[])
3215 int main(int argc, const char *argv[])
3219 unsigned int random_seed;
3221 struct passwd *pw = NULL;
3222 struct group *grp = NULL;
3225 char *pre_chroot_nslookup_to_load_resolver = NULL;
3232 #if !defined(_WIN32)
3240 * Parse the command line arguments
3242 * XXX: simply printing usage information in case of
3243 * invalid arguments isn't particularly user friendly.
3245 while (++argc_pos < argc)
3248 /* Check to see if the service must be installed or uninstalled */
3249 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3251 const char *pName = argv[argc_pos] + 9;
3254 exit( (install_service(pName)) ? 0 : 1 );
3256 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3258 const char *pName = argv[argc_pos] + 11;
3261 exit((uninstall_service(pName)) ? 0 : 1);
3263 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3265 bRunAsService = TRUE;
3266 w32_set_service_cwd();
3267 atexit(w32_service_exit_notify);
3270 #endif /* defined(_WIN32) */
3273 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3275 if (strcmp(argv[argc_pos], "--help") == 0)
3280 else if(strcmp(argv[argc_pos], "--version") == 0)
3282 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3288 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3293 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3295 if (++argc_pos == argc) usage(argv[0]);
3296 pidfile = strdup(argv[argc_pos]);
3299 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3301 if (++argc_pos == argc) usage(argv[argc_pos]);
3303 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3306 if (NULL == (grp = getgrnam(p)))
3308 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3312 if (NULL == (pw = getpwnam(argv[argc_pos])))
3314 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3317 if (p != NULL) *--p = '\0';
3320 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3322 if (++argc_pos == argc) usage(argv[0]);
3323 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3326 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3330 #endif /* defined(unix) */
3332 else if (argc_pos + 1 != argc)
3335 * This is neither the last command line
3336 * option, nor was it recognized before,
3337 * therefore it must be invalid.
3343 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3345 configfile = argv[argc_pos];
3348 } /* -END- while (more arguments) */
3351 if ( *configfile != '/' )
3353 char cwd[BUFFER_SIZE];
3355 size_t abs_file_size;
3357 /* make config-filename absolute here */
3358 if (NULL == getcwd(cwd, sizeof(cwd)))
3360 perror("failed to get current working directory");
3365 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3366 basedir = strdup(cwd);
3368 if (NULL == basedir ||
3369 NULL == (abs_file = malloc(abs_file_size)))
3371 perror("malloc failed");
3374 strlcpy(abs_file, basedir, abs_file_size);
3375 strlcat(abs_file, "/", abs_file_size );
3376 strlcat(abs_file, configfile, abs_file_size);
3377 configfile = abs_file;
3379 #endif /* defined unix */
3383 clients->next = NULL;
3385 /* XXX: factor out initialising after the next stable release. */
3388 #elif defined(_WIN32)
3392 /* Prepare mutexes if supported and necessary. */
3393 initialize_mutexes();
3395 /* Enable logging until further notice. */
3396 init_log_module(Argv[0]);
3398 random_seed = (unsigned int)time(NULL);
3400 srandom(random_seed);
3403 #endif /* ifdef HAVE_RANDOM */
3406 * Unix signal handling
3408 * Catch the abort, interrupt and terminate signals for a graceful exit
3409 * Catch the hangup signal so the errlog can be reopened.
3410 * Ignore the broken pipe signals (FIXME: Why?)
3412 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3415 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3416 const int ignored_signals[] = { SIGPIPE, 0 };
3418 for (idx = 0; catched_signals[idx] != 0; idx++)
3420 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3421 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3423 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3424 #endif /* ifdef sun */
3426 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3430 for (idx = 0; ignored_signals[idx] != 0; idx++)
3432 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3434 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3439 #else /* ifdef _WIN32 */
3440 # ifdef _WIN_CONSOLE
3442 * We *are* in a windows console app.
3443 * Print a verbose messages about FAQ's and such
3445 printf("%s", win32_blurb);
3446 # endif /* def _WIN_CONSOLE */
3447 #endif /* def _WIN32 */
3450 /* Initialize the CGI subsystem */
3451 cgi_init_error_messages();
3454 * If runnig on unix and without the --nodaemon
3455 * option, become a daemon. I.e. fork, detach
3456 * from tty and get process group leadership
3469 if ( pid < 0 ) /* error */
3474 else if ( pid != 0 ) /* parent */
3479 * must check for errors
3480 * child died due to missing files aso
3483 wpid = waitpid( pid, &status, WNOHANG );
3492 /* Should be more portable, but not as well tested */
3497 #else /* ndef __FreeBSD__ */
3499 #endif /* ndef __FreeBSD__ */
3500 fd = open("/dev/tty", O_RDONLY);
3503 /* no error check here */
3504 ioctl( fd, TIOCNOTTY,0 );
3509 * stderr (fd 2) will be closed later on, when the
3510 * log file has been parsed.
3517 } /* -END- if (!no_daemon) */
3520 * As soon as we have written the PID file, we can switch
3521 * to the user and group ID indicated by the --user option
3527 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3529 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3535 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3537 /* Read the time zone file from /etc before doing chroot. */
3539 if (NULL != pre_chroot_nslookup_to_load_resolver
3540 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3542 /* Initialize resolver library. */
3543 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3545 if (chroot(pw->pw_dir) < 0)
3547 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3551 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3554 if (setuid(pw->pw_uid))
3556 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3560 char putenv_dummy[64];
3562 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3563 if (putenv(putenv_dummy) != 0)
3565 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3568 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3569 if (putenv(putenv_dummy) != 0)
3571 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3577 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3580 #endif /* defined unix */
3583 /* This will be FALSE unless the command line specified --service
3587 /* Yup, so now we must attempt to establish a connection
3588 * with the service dispatcher. This will only work if this
3589 * process was launched by the service control manager to
3590 * actually run as a service. If this isn't the case, i've
3591 * known it take around 30 seconds or so for the call to return.
3594 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3595 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3597 /* Service has run, and at this point is now being stopped, so just return */
3602 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3604 /* An error occurred. Usually it's because --service was wrongly specified
3605 * and we were unable to connect to the Service Control Dispatcher because
3606 * it wasn't expecting us and is therefore not listening.
3608 * For now, just continue below to call the listen_loop function.
3611 #endif /* def _WIN32 */
3621 /*********************************************************************
3623 * Function : bind_port_helper
3625 * Description : Bind the listen port. Handles logging, and aborts
3629 * 1 : config = Privoxy configuration. Specifies port
3632 * Returns : Port that was opened.
3634 *********************************************************************/
3635 static jb_socket bind_port_helper(struct configuration_spec * config)
3640 if (config->haddr == NULL)
3642 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3647 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3648 config->hport, config->haddr);
3651 result = bind_port(config->haddr, config->hport, &bfd);
3658 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3659 "There may be another Privoxy or some other "
3660 "proxy running on port %d",
3661 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3662 config->hport, config->hport);
3665 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3666 "The hostname is not resolvable",
3667 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3670 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3671 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3674 /* shouldn't get here */
3675 return JB_INVALID_SOCKET;
3678 config->need_bind = 0;
3685 /* Without this simple workaround we get this compiler warning from _beginthread
3686 * warning C4028: formal parameter 1 different from declaration
3688 void w32_service_listen_loop(void *p)
3692 #endif /* def _WIN32 */
3695 /*********************************************************************
3697 * Function : listen_loop
3699 * Description : bind the listen port and enter a "FOREVER" listening loop.
3705 *********************************************************************/
3706 static void listen_loop(void)
3708 struct client_state *csp = NULL;
3710 struct configuration_spec * config;
3712 config = load_config();
3714 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3716 * XXX: Should be relocated once it no
3717 * longer needs to emit log messages.
3719 initialize_reusable_connections();
3720 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3722 bfd = bind_port_helper(config);
3724 #ifdef FEATURE_GRACEFUL_TERMINATION
3725 while (!g_terminate)
3730 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3731 while (waitpid(-1, NULL, WNOHANG) > 0)
3733 /* zombie children */
3735 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3738 * Free data that was used by died threads
3744 * Re-open the errlog after HUP signal
3746 if (received_hup_signal)
3748 init_error_log(Argv[0], config->logfile);
3749 received_hup_signal = 0;
3753 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3755 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3759 csp->flags |= CSP_FLAG_ACTIVE;
3760 csp->sfd = JB_INVALID_SOCKET;
3762 csp->config = config = load_config();
3764 if ( config->need_bind )
3767 * Since we were listening to the "old port", we will not see
3768 * a "listen" param change until the next IJB request. So, at
3769 * least 1 more request must be made for us to find the new
3770 * setting. I am simply closing the old socket and binding the
3773 * Which-ever is correct, we will serve 1 more page via the
3774 * old settings. This should probably be a "show-proxy-args"
3775 * request. This should not be a so common of an operation
3776 * that this will hurt people's feelings.
3781 bfd = bind_port_helper(config);
3784 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3786 if (!accept_connection(csp, bfd))
3788 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3801 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3804 #ifdef FEATURE_TOGGLE
3805 if (global_toggle_state)
3806 #endif /* def FEATURE_TOGGLE */
3808 csp->flags |= CSP_FLAG_TOGGLED_ON;
3811 if (run_loader(csp))
3813 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3814 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3818 if (block_acl(NULL,csp))
3820 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3821 close_socket(csp->cfd);
3825 #endif /* def FEATURE_ACL */
3827 /* add it to the list of clients */
3828 csp->next = clients->next;
3829 clients->next = csp;
3831 if (config->multi_threaded)
3835 /* this is a switch () statment in the C preprocessor - ugh */
3836 #undef SELECTED_ONE_OPTION
3838 /* Use Pthreads in preference to native code */
3839 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3840 #define SELECTED_ONE_OPTION
3842 pthread_t the_thread;
3843 pthread_attr_t attrs;
3845 pthread_attr_init(&attrs);
3846 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3847 errno = pthread_create(&the_thread, &attrs,
3848 (void * (*)(void *))serve, csp);
3849 child_id = errno ? -1 : 0;
3850 pthread_attr_destroy(&attrs);
3854 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3855 #define SELECTED_ONE_OPTION
3856 child_id = _beginthread(
3857 (void (*)(void *))serve,
3862 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3863 #define SELECTED_ONE_OPTION
3864 child_id = _beginthread(
3865 (void(* _Optlink)(void*))serve,
3871 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3872 #define SELECTED_ONE_OPTION
3874 thread_id tid = spawn_thread
3875 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3877 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3879 child_id = (int) tid;
3888 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3889 #define SELECTED_ONE_OPTION
3890 csp->cfd = ReleaseSocket(csp->cfd, -1);
3893 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3894 NP_Output, Output(),
3895 NP_CloseOutput, FALSE,
3896 NP_Name, (ULONG)"privoxy child",
3900 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3901 NP_Output, Output(),
3902 NP_CloseOutput, FALSE,
3903 NP_Name, (ULONG)"privoxy child",
3904 NP_StackSize, 200*1024,
3910 ((struct Task *)child_id)->tc_UserData = csp;
3911 Signal((struct Task *)child_id, SIGF_SINGLE);
3916 #if !defined(SELECTED_ONE_OPTION)
3919 /* This block is only needed when using fork().
3920 * When using threads, the server thread was
3921 * created and run by the call to _beginthread().
3923 if (child_id == 0) /* child */
3926 #ifdef FEATURE_TOGGLE
3927 int inherited_toggle_state = global_toggle_state;
3928 #endif /* def FEATURE_TOGGLE */
3933 * If we've been toggled or we've blocked the request, tell Mom
3936 #ifdef FEATURE_TOGGLE
3937 if (inherited_toggle_state != global_toggle_state)
3939 rc |= RC_FLAG_TOGGLED;
3941 #endif /* def FEATURE_TOGGLE */
3943 #ifdef FEATURE_STATISTICS
3944 if (csp->flags & CSP_FLAG_REJECTED)
3946 rc |= RC_FLAG_BLOCKED;
3948 #endif /* ndef FEATURE_STATISTICS */
3952 else if (child_id > 0) /* parent */
3954 /* in a fork()'d environment, the parent's
3955 * copy of the client socket and the CSP
3959 #if !defined(_WIN32) && !defined(__CYGWIN__)
3961 wait( &child_status );
3964 * Evaluate child's return code: If the child has
3965 * - been toggled, toggle ourselves
3966 * - blocked its request, bump up the stats counter
3969 #ifdef FEATURE_TOGGLE
3970 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3972 global_toggle_state = !global_toggle_state;
3974 #endif /* def FEATURE_TOGGLE */
3976 #ifdef FEATURE_STATISTICS
3978 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3982 #endif /* def FEATURE_STATISTICS */
3984 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3985 close_socket(csp->cfd);
3986 csp->flags &= ~CSP_FLAG_ACTIVE;
3990 #undef SELECTED_ONE_OPTION
3991 /* end of cpp switch () */
3993 if (child_id < 0) /* failed */
3995 char buf[BUFFER_SIZE];
3997 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3999 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
4001 write_socket(csp->cfd, buf, strlen(buf));
4002 close_socket(csp->cfd);
4003 csp->flags &= ~CSP_FLAG_ACTIVE;
4014 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4016 /* Clean up. Aim: free all memory (no leaks) */
4017 #ifdef FEATURE_GRACEFUL_TERMINATION
4019 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4021 unload_current_config_file();
4022 unload_current_actions_file();
4023 unload_current_re_filterfile();
4024 #ifdef FEATURE_TRUST
4025 unload_current_trust_file();
4028 if (config->multi_threaded)
4035 } while ((clients->next != NULL) && (--i > 0));
4039 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4050 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4051 /* Cleanup - remove taskbar icon etc. */
4056 #endif /* FEATURE_GRACEFUL_TERMINATION */