1 const char jcc_rcs[] = "$Id: jcc.c,v 1.186 2008/09/04 08:13:58 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 1.186 2008/09/04 08:13:58 fabiankeil
37 * Prepare for critical sections on Windows by adding a
38 * layer of indirection before the pthread mutex functions.
40 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
41 * Remove FEATURE_COOKIE_JAR.
43 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
44 * - Silence LLVM/Clang complaint.
45 * - Make received_hup_signal static.
46 * - Hide definitions for basedir, pidfile and received_hup_signal
47 * from __EMX__ as they only seem to be used in case of #ifdef unix.
49 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
50 * Accept Shoutcast responses again. Problem reported
51 * and fix suggested by Stefan in #2062860.
53 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
54 * Fix possible NULL-pointer dereference reported
55 * by din_a4 in #2003937. Pointy hat to me.
57 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
58 * Streamline sed()'s prototype and declare
59 * the header parse and add structures static.
61 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
62 * - Mark csp as immutable for send_crunch_response().
63 * - Fix comment spelling.
65 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
66 * Factor update_server_headers() out of sed(), ditch the
67 * first_run hack and make server_patterns_light static.
69 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
70 * Don't provide get_header() with the whole client state
71 * structure when it only needs access to csp->iob.
73 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
74 * Make the "read the rest of the headers" loop a bit more readable.
76 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
77 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
78 * - Remove useless comment.
80 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
81 * Fix comment grammar.
83 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
84 * Remove the pointless buffer in client_protocol_is_unsupported().
86 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
87 * Least-effort fix for bug #1821930 (reported by Lee):
88 * If the response doesn't look like HTTP,
89 * tell the client and log the problem.
91 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
92 * Don't pass the whole csp structure to flush_socket()
93 * when it only needs a file descriptor and a buffer.
95 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
96 * Remove kill-popups action.
98 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
99 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
101 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
102 * Remove the treat-forbidden-connects-like-blocks action. We now
103 * use the "blocked" page for forbidden CONNECT requests by default.
105 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
106 * Also use shiny new connect_port_is_forbidden() in jcc.c.
108 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
109 * Rename url_actions() to get_url_actions() and let it
110 * use the standard parameter ordering.
112 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
113 * Let forward_url() use the standard parameter ordering
114 * and mark its second parameter immutable.
116 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
117 * Remove the "Listening ... for local connections only" log message.
118 * Whether or not remote connections are able to reach Privoxy is up
119 * to the operating system.
121 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
122 * Prevent the log messages for CONNECT requests to unacceptable
123 * ports from printing the limit-connect argument as [null] if
124 * limit-connect hasn't been explicitly enabled.
126 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
127 * Make sure all console-mode apps get a usage() instance
129 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
130 * Reword NO_SERVER_DATA_RESPONSE to make it harder
131 * to misunderstand what the message is all about.
133 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
134 * Unbreak trustfile which previously didn't work without
135 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
137 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
138 * Plug memory leak. Spotted by Valgrind, triggered by
139 * Privoxy-Regression-Test feeding proxyfuzz.py.
141 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
142 * In the HTTP snipplets, refer to the client as client.
144 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
145 * Emit a log message when activating the MS IIS5 hack.
147 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
148 * Log the "weak randomization factor" warning only
149 * once for mingw32 and provide some more details.
151 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
152 * Initialize log module after initializing mutexes, future
153 * deadlocks in that code should now work cross-platform.
155 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
156 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
157 * Reported by Bert van Leeuwen in BR#1818808.
159 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
160 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
162 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
163 * When in daemon mode, close stderr after the configuration file has been
164 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
166 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
167 * - Fix a crash when parsing invalid requests whose first header
168 * is rejected by get_header(). Regression (re?)introduced
169 * in r1.143 by yours truly.
170 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
171 * client_cookie_adder() to make sure send-vanilla-wafer can be
172 * controlled through tags (and thus regression-tested).
174 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
175 * - Move get_filter_function() from jcc.c to filters.c
176 * so the filter functions can be static.
177 * - Don't bother filtering body-less responses.
179 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
180 * Execute content filters through execute_content_filter().
182 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
183 * Initialize req to NULL to make sure it's defined if the
184 * first read_socket() call fails. Reported by icmp30.
186 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
187 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
188 * extensive comments moved to user manual.
190 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
191 * Don't crash if a broken header filter wiped out the request line.
193 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
194 * Fix byte_count calculation in case of flushes
195 * and don't parse the server headers a second time.
197 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
198 * - If there's a connection problem after we already forwarded
199 * parts of the original content, just hang up. Fixes BR#1776724.
200 * - Fix warnings about unused code on mingw32.
201 * - In case of flushes, calculate the byte count
202 * less incorrectly (I think).
204 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
205 * Add some more prototypes for static functions.
207 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
208 * Comment out request_contains_null_bytes() until it's used again.
210 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
211 * #1763173 from Stefan Huehner: s@const static@static const@
212 * and declare some more functions static.
214 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
215 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
216 * and explain why they were rejected in the first place.
217 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
218 * CONNECT requests. The request line was missing.
219 * - Add two more XXX reminders as we don't have enough already.
221 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
222 * As Hal noticed, checking dispatch_cgi() as the last cruncher
223 * looks like a bug if CGI requests are blocked unintentionally,
224 * so don't do it unless the user enabled the new config option
225 * "allow-cgi-request-crunching".
227 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
228 * - Allow to rewrite the request destination behind the client's back.
229 * - Turn the weird-looking unconditional for loop that
230 * reads the client request into a conditional while loop.
231 * Move the stuff that only runs once out of the loop.
232 * - Move parts of chat(), server_content_type() and the
233 * necessary stuff to fix BR#1750917 into get_filter_function().
235 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
236 * Temporary workaround for BR#1730105.
238 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
239 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
240 * deadlocks and crashes on OpenBSD and possibly other OS with neither
241 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
242 * Thanks to Ralf Horstmann for report and solution.
244 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
245 * Add forward-override{} to change the forwarding settings through
246 * action sections. This is mainly interesting to forward different
247 * clients differently (for example based on User-Agent or request
250 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
251 * - Let usage() mention the --chroot parameter.
252 * - Use read_socket() consistently and always leave
253 * the last buffer byte alone, even in cases where
254 * null termination (currently) doesn't matter.
256 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
257 * - Fix config file loading on Unix if no config file is specified.
258 * Since r1.97 Privoxy would always interpret the last argument as
259 * config file, even if it's a valid command line option.
260 * - Abort in case of unrecognized command line options. Closes #1719696.
261 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
262 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
264 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
265 * - Don't rerun crunchers that only depend on the request URL.
266 * - Don't count redirects and CGI requests as "blocked requests".
268 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
269 * Support crunching based on tags created by server-header taggers.
271 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
272 * Make HTTP snippets static (again). Add a Content-Type for those
273 * with content so the browser doesn't guess it based on the URL.
275 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
276 * Move crunching and request line rebuilding out of chat().
278 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
279 * Introduce tags as alternative way to specify which
280 * actions apply to a request. At the moment tags can be
281 * created based on client and server headers.
283 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
284 * Don't CLF-log CONNECT requests twice.
286 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
287 * Log the source address for ACL-related connection drops.
289 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
290 * New config option: enforce-blocks.
292 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
293 * - Move null byte check into separate function.
294 * - Don't confuse the client with error pages
295 * if a CONNECT request was already confirmed.
297 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
298 * Speed up NULL byte escaping and only log the complete
299 * NULL byte requests with header debugging enabled.
301 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
302 * Answer requests that contain NULL bytes with
303 * a custom response instead of waiting for more
304 * data until the client eventually hangs up.
306 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
307 * - Move delivery and logging of crunched responses
308 * from chat() into send_crunch_response().
309 * - Display the reason for generating http_responses.
310 * - Log the content length for LOG_LEVEL_CLF correctly
311 * (still incorrect for some fixed responses).
312 * - Reword an incorrect comment about
313 * treat-forbidden-connects-like-blocks violating
315 * - Add some log messages.
317 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
318 * Move mutex initialization into separate
319 * function and exit in case of errors.
321 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
322 * - Start to reduce chat()'s line count and move
323 * parts of it into separate functions.
324 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
326 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
327 * - Add Proxy-Agent header to HTTP snippets that are
328 * supposed to reach HTTP clients only.
329 * - Made a few CONNECT log messages more descriptive.
330 * - Catch completely empty server responses (as seen
331 * with Tor's fake ".noconnect" top level domain).
332 * - Use shiny new "forwarding-failed" template for socks errors.
334 * Revision 1.118 2007/01/07 07:43:43 joergs
335 * AmigaOS4 support added.
337 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
338 * Added config option accept-intercepted-requests
339 * and disabled it by default.
341 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
342 * Reverted parts of my last commit
343 * to keep error handling working.
345 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
346 * Fixed gcc43 conversion warnings.
348 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
349 * Fix -pedantic ISO C warning about converting
350 * from function pointer to object pointer.
352 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
353 * Silence compiler warning I introduced with my last commit.
355 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
356 * Mutex protect rand() if POSIX threading
357 * is used, warn the user if that's not possible
358 * and stop using it on _WIN32 where it could
361 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
362 * Don't prevent core dumps by catching SIGABRT.
363 * It's rude and makes debugging unreasonable painful.
365 * Revision 1.110 2006/12/13 14:52:53 etresoft
366 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
368 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
369 * Privoxy is now able to run as intercepting
370 * proxy in combination with any packet filter
371 * that does the port redirection. The destination
372 * is extracted from the "Host:" header which
373 * should be available for nearly all requests.
375 * Moved HTTP snipplets into jcc.c.
376 * Added error message for gopher proxy requests.
378 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
379 * Only unlink the pidfile if it's actually used.
381 * Change order of interception checks to make
382 * it possible to block or redirect requests for
385 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
386 * Make pthread mutex locking more generic. Instead of
387 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
388 * and use mutex locking unless there is an _r function
389 * available. Better safe than sorry.
391 * Fixes "./configure --disable-pthread" and should result
392 * in less threading-related problems on pthread-using platforms,
393 * but it still doesn't fix BR#1122404.
395 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
396 * Move pthread.h inclusion from jcc.c to jcc.h.
397 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
399 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
400 * Don't exit after receiving the second SIGHUP on Solaris.
402 * Fixes BR 1052235, but the same problem may exist on other
403 * systems. Once 3.0.6 is out we should use sigset()
404 * where available and see if it breaks anything.
406 * Revision 1.104 2006/09/23 13:26:38 roro
407 * Replace TABs by spaces in source code.
409 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
410 * Fix +redirect{}. Didn't work with -fast-redirects.
412 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
413 * Respond with 400 and a short text message
414 * if the client tries to use Privoxy as FTP proxy.
416 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
417 * Make number of retries in case of forwarded-connect problems
418 * a config file option (forwarded-connect-retries) and use 0 as
421 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
422 * Set random(3) seed.
424 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
425 * Follow the OpenBSD port's lead and protect the resolve
426 * functions on OpenBSD as well.
428 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
429 * --user fix. Only use the user as group if no group is specified.
430 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
432 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
433 * Windows service (re-)integration
437 * --install[:service_name]
438 * --uninstall[:service_name]
441 * They work as follows:
442 * --install will create a service for you and then terminate.
443 * By default the service name will be "privoxy" (without the quotes).
444 * However you can run multiple services if you wish, just by adding
445 * a colon and then a name (no spaces).
447 * --uninstall follows the exact same rules a --install.
449 * --service is used when the program is executed by the service
450 * control manager, and in normal circumstances would never be
451 * used as a command line argument.
453 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
454 * Windows service integration
456 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
457 * Incorporate Fabian Keil's patch work:
458 http://www.fabiankeil.de/sourcecode/privoxy/
460 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
461 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
462 * with what was really the latest development (the v_3_0_branch branch)
464 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
465 * Thanks to Jindrich Makovicka for a race condition fix for the log
466 * file. The race condition remains for non-pthread implementations.
467 * Reference patch #1175720.
469 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
470 * Add the ability to check jpeg images for invalid
471 * lengths of comment blocks. Defensive strategy
472 * against the exploit:
473 * Microsoft Security Bulletin MS04-028
474 * Buffer Overrun in JPEG Processing (GDI+) Could
475 * Allow Code Execution (833987)
476 * Enabled with +inspect-jpegs in actions files.
478 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
479 * - Fixed usage info for non-unix platforms
480 * - Fixed small cmdline parsing bug
482 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
483 * Diagnostics: Now preserve the returncode of pthread_create
484 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
486 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
487 * No longer ignore SIGCHLD. Fixes bug #769381
489 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
490 * Close jarfile on graceful exit, remove stray line
492 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
493 * Cosmetics: Killed a warning, a typo and an allocation left at exit
495 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
496 * No longer rely on non-POSIX.1 extensions of getcwd().
499 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
500 * Replaced setenv() by posix-compliant putenv()
501 * Thanks to Neil McCalden (nmcc AT users.sf.net).
503 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
504 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
506 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
507 * Clean-up and extension of improvements for forked mode:
508 * - Child's return code now consists of flags RC_FLAG_*
509 * - Reporting toggle to parent now properly #ifdef'ed
510 * - Children now report blocking to parent. This enables
511 * statistics in forked mode
513 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
514 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
515 * Children now report having been toggled through _exit(17), parents
516 * watch for that code and toggle themselves if found.
518 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
519 * Wrapping all *_r functions (the non-_r versions of them) with
520 * mutex semaphores for OSX. Hopefully this will take care of all
521 * of those pesky crash reports.
523 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
524 * Fixed two mostly harmless mem leaks
526 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
527 * Fix: Head of global clients list now initialized to NULL
529 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
530 * Added basic support for OPTIONS and TRACE HTTP methods:
531 * - New interceptor direct_response() added in chat().
532 * - sed() moved to earlier in the process, so that the
533 * Host: header is evaluated before actions and forwarding
536 * Revision 1.92 2002/05/08 16:00:46 oes
537 * Chat's buffer handling:
538 * - Fixed bug with unchecked out-of-mem conditions
539 * while reading client request & server headers
540 * - No longer predict if the buffer limit will be exceeded
541 * in the next read -- check add_to_iob's new
542 * return code. If buffer couldn't be extended
543 * (policy or out-of-mem) while
544 * - reading from client: abort
545 * - reading server headers: send error page
546 * - buffering server body for filter: flush,
547 * and if that fails: send error page
549 * Revision 1.91 2002/04/08 20:35:58 swa
552 * Revision 1.90 2002/04/02 14:57:28 oes
553 * Made sending wafers independent of FEATURE_COOKIE_JAR
555 * Revision 1.89 2002/03/31 17:18:59 jongfoster
556 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
558 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
559 * More compiler warning message maintenance
561 * Revision 1.87 2002/03/26 22:29:54 swa
562 * we have a new homepage!
564 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
565 * Workaround for closing the jarfile before load_config() comes around again
567 * Revision 1.85 2002/03/24 15:23:33 jongfoster
570 * Revision 1.84 2002/03/24 13:25:43 swa
571 * name change related issues
573 * Revision 1.83 2002/03/16 23:54:06 jongfoster
574 * Adding graceful termination feature, to help look for memory leaks.
575 * If you enable this (which, by design, has to be done by hand
576 * editing config.h) and then go to http://i.j.b/die, then the program
577 * will exit cleanly after the *next* request. It should free all the
578 * memory that was used.
580 * Revision 1.82 2002/03/13 00:27:05 jongfoster
583 * Revision 1.81 2002/03/12 01:42:50 oes
584 * Introduced modular filters
586 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
587 * OS/2 port maintenance:
588 * - Fixed EMX build - it had decayed a little
589 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
590 * substituted a memset for now.
592 * Revision 1.79 2002/03/09 20:03:52 jongfoster
593 * - Making various functions return int rather than size_t.
594 * (Undoing a recent change). Since size_t is unsigned on
595 * Windows, functions like read_socket that return -1 on
596 * error cannot return a size_t.
598 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
599 * crashes, and also frequently caused JB to jump to 100%
600 * CPU and stay there. (Because it thought it had just
601 * read ((unsigned)-1) == 4Gb of data...)
603 * - The signature of write_socket has changed, it now simply
604 * returns success=0/failure=nonzero.
606 * - Trying to get rid of a few warnings --with-debug on
607 * Windows, I've introduced a new type "jb_socket". This is
608 * used for the socket file descriptors. On Windows, this
609 * is SOCKET (a typedef for unsigned). Everywhere else, it's
610 * an int. The error value can't be -1 any more, so it's
611 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
612 * Windows it maps to the #define INVALID_SOCKET.)
614 * - The signature of bind_port has changed.
616 * Revision 1.78 2002/03/08 21:35:04 oes
617 * Added optional group supplement to --user option. Will now use default group of user if no group given
619 * Revision 1.77 2002/03/07 03:52:06 oes
620 * - Fixed compiler warnings etc
621 * - Improved handling of failed DNS lookups
623 * Revision 1.76 2002/03/06 22:54:35 jongfoster
624 * Automated function-comment nitpicking.
626 * Revision 1.75 2002/03/06 10:02:19 oes
627 * Fixed stupid bug when --user was not given
629 * Revision 1.74 2002/03/06 00:49:31 jongfoster
630 * Fixing warning on Windows
631 * Making #ifdefs that refer to the same variable consistently
632 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
634 * Revision 1.73 2002/03/05 23:57:30 hal9
635 * Stray character 's' on line 1618 was breaking build.
637 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
638 * - Re-enable OS/2 building after new parms were added
639 * - Fix false out of memory report when resolving CGI templates when no IP
640 * address is available of failed attempt (a la no such domain)
642 * Revision 1.71 2002/03/05 18:13:56 oes
643 * Added --user option
645 * Revision 1.70 2002/03/05 04:52:42 oes
646 * Deleted non-errlog debugging code
648 * Revision 1.69 2002/03/04 23:50:00 jongfoster
649 * Splitting off bind_port() call into bind_port_helper(), with
652 * Revision 1.68 2002/03/04 20:17:32 oes
655 * Revision 1.67 2002/03/04 18:18:57 oes
656 * - Removed _DEBUG mode
657 * - Cleand up cmdline parsing
658 * - Introduced --no-daemon, --pidfile options
659 * - Cleaned up signal handling:
660 * - Terminate cleanly on INT, TERM and ABRT
661 * - Schedule logfile for re-opening on HUP
662 * - Ignore CHLD and PIPE
663 * - Leave the rest with their default handlers
664 * - Uniform handler registration
665 * - Added usage() function
666 * - Played styleguide police
668 * Revision 1.66 2002/03/03 15:06:55 oes
669 * Re-enabled automatic config reloading
671 * Revision 1.65 2002/03/03 14:49:11 oes
672 * Fixed CLF logging: Now uses client's original HTTP request
674 * Revision 1.64 2002/03/03 09:18:03 joergs
675 * Made jumbjuster work on AmigaOS again.
677 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
678 * Clean up a little CRLF unpleasantness that suddenly appeared
680 * Revision 1.62 2002/02/20 23:17:23 jongfoster
681 * Detecting some out-of memory conditions and exiting with a log message.
683 * Revision 1.61 2002/01/17 21:01:52 jongfoster
684 * Moving all our URL and URL pattern parsing code to urlmatch.c.
686 * Revision 1.60 2001/12/30 14:07:32 steudten
687 * - Add signal handling (unix)
688 * - Add SIGHUP handler (unix)
689 * - Add creation of pidfile (unix)
690 * - Add action 'top' in rc file (RH)
691 * - Add entry 'SIGNALS' to manpage
692 * - Add exit message to logfile (unix)
694 * Revision 1.59 2001/12/13 14:07:18 oes
695 * Fixed Bug: 503 error page now sent OK
697 * Revision 1.58 2001/11/30 23:37:24 jongfoster
698 * Renaming the Win32 config file to config.txt - this is almost the
699 * same as the corresponding UNIX name "config"
701 * Revision 1.57 2001/11/16 00:47:43 jongfoster
702 * Changing the tty-disconnection code to use setsid().
704 * Revision 1.56 2001/11/13 20:20:54 jongfoster
705 * Tabs->spaces, fixing a bug with missing {} around an if()
707 * Revision 1.55 2001/11/13 20:14:53 jongfoster
708 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
710 * Revision 1.54 2001/11/07 00:03:14 steudten
711 * Give reliable return value if an error
712 * occurs not just 0 with new daemon mode.
714 * Revision 1.53 2001/11/05 21:41:43 steudten
715 * Add changes to be a real daemon just for unix os.
716 * (change cwd to /, detach from controlling tty, set
717 * process group and session leader to the own process.
719 * Add some fatal-error log message for failed malloc().
720 * Add '-d' if compiled with 'configure --with-debug' to
721 * enable debug output.
723 * Revision 1.52 2001/10/26 20:11:20 jongfoster
724 * Fixing type mismatch
726 * Revision 1.51 2001/10/26 17:38:28 oes
729 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
730 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
731 * threads to call select() simultaneously. So, it's time to do a real, live,
732 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
733 * (native). Both versions will work, but using __OS2__ offers multi-threading.
735 * Revision 1.49 2001/10/23 21:41:35 jongfoster
736 * Added call to initialize the (statically-allocated of course)
737 * "out of memory" CGI response.
739 * Revision 1.48 2001/10/10 19:56:46 jongfoster
740 * Moving some code that wasn't cookie-related out of an #ifdef
743 * Revision 1.47 2001/10/10 16:44:36 oes
744 * Added CONNECT destination port limitation check
746 * Revision 1.46 2001/10/08 15:17:41 oes
747 * Re-enabled SSL forwarding
749 * Revision 1.45 2001/10/07 15:42:11 oes
750 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
752 * Moved downgrading of the HTTP version from parse_http_request to
753 * chat(), since we can't decide if it is necessary before we have
754 * determined the actions for the URL. The HTTP command is now
755 * *always* re-built so the repairs need no longer be special-cased.
757 * filter_popups now gets a csp pointer so it can raise the new
758 * CSP_FLAG_MODIFIED flag.
762 * Added configurable size limit for the IOB. If the IOB grows so
763 * large that the next read would exceed the limit, the header
764 * is generated, and the header & unfiltered buffer are flushed
765 * to the client. Chat then continues in non-buffering,
766 * non-filtering body mode.
768 * Revision 1.44 2001/10/02 18:13:57 oes
771 * Revision 1.43 2001/10/02 15:32:13 oes
772 * Moved generation of hdr
774 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
775 * Cleaning up 2 compiler warnings on OS/2.
777 * Revision 1.41 2001/09/16 17:05:14 jongfoster
778 * Removing unused #include showarg.h
780 * Revision 1.40 2001/09/16 15:41:45 jongfoster
781 * Fixing signed/unsigned comparison warning.
783 * Revision 1.39 2001/09/16 13:21:27 jongfoster
784 * Changes to use new list functions.
786 * Revision 1.38 2001/09/16 13:01:46 jongfoster
787 * Removing redundant function call that zeroed zalloc()'d memory.
789 * Revision 1.37 2001/09/10 11:12:24 oes
790 * Deleted unused variable
792 * Revision 1.36 2001/09/10 10:56:15 oes
793 * Silenced compiler warnings
795 * Revision 1.35 2001/07/31 14:44:22 oes
796 * Deleted unused size parameter from filter_popups()
798 * Revision 1.34 2001/07/30 22:08:36 jongfoster
799 * Tidying up #defines:
800 * - All feature #defines are now of the form FEATURE_xxx
801 * - Permanently turned off WIN_GUI_EDIT
802 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
804 * Revision 1.33 2001/07/29 19:32:00 jongfoster
805 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
807 * Revision 1.32 2001/07/29 18:47:05 jongfoster
808 * Adding missing #include "loadcfg.h"
810 * Revision 1.31 2001/07/29 12:17:48 oes
811 * Applied pthread fix by Paul Lieverse
813 * Revision 1.30 2001/07/25 22:57:13 jongfoster
814 * __BEOS__ no longer overrides FEATURE_PTHREAD.
815 * This is because FEATURE_PTHREAD will soon be widely used, so I
816 * want to keep it simple.
818 * Revision 1.29 2001/07/24 12:47:06 oes
819 * Applied BeOS support update by Eugenia
821 * Revision 1.28 2001/07/23 13:26:12 oes
822 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
824 * Revision 1.27 2001/07/19 19:09:47 haroon
825 * - Added code to take care of the situation where while processing the first
826 * server response (which includes the server header), after finding the end
827 * of the headers we were not looking past the end of the headers for
828 * content modification. I enabled it for filter_popups.
829 * Someone else should look to see if other similar operations should be
830 * done to the discarded portion of the buffer.
832 * Note 2001/07/20: No, the other content modification mechanisms will process
833 * the whole iob later anyway. --oes
835 * Revision 1.26 2001/07/18 12:31:36 oes
838 * Revision 1.25 2001/07/15 19:43:49 jongfoster
839 * Supports POSIX threads.
840 * Also removed some unused #includes.
842 * Revision 1.24 2001/07/13 14:00:40 oes
843 * - Generic content modification scheme:
844 * Each feature has its own applicability flag that is set
845 * from csp->action->flags.
846 * Replaced the "filtering" int flag , by a function pointer
847 * "content_filter" to the function that will do the content
848 * modification. If it is != NULL, the document will be buffered
849 * and processed through *content_filter, which must set
850 * csp->content_length and return a modified copy of the body
851 * or return NULL (on failiure).
852 * - Changed csp->is_text to the more generic bitmap csp->content_type
853 * which can currently take the valued CT_TEXT or CT_GIF
855 * - Removed all #ifdef PCRS
857 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
858 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
859 * the `block_acl' call. This prevents a compilation error when the user
860 * does not wish to use the "ACL" feature.
862 * Revision 1.22 2001/06/29 21:45:41 oes
863 * Indentation, CRLF->LF, Tab-> Space
865 * Revision 1.21 2001/06/29 13:29:36 oes
866 * - Cleaned up, improved comments
867 * - Unified all possible interceptors (CGI,
868 * block, trust, fast_redirect) in one
869 * place, with one (CGI) answer generation
870 * mechansim. Much clearer now.
871 * - Removed the GIF image generation, which
872 * is now done in filters.c:block_url()
873 * - Made error conditions like domain lookup
874 * failiure or (various) problems while talking
875 * to the server use cgi.c:error_response()
876 * instead of generating HTML/HTTP in chat() (yuck!)
877 * - Removed logentry from cancelled commit
879 * Revision 1.20 2001/06/09 10:55:28 jongfoster
880 * Changing BUFSIZ ==> BUFFER_SIZE
882 * Revision 1.19 2001/06/07 23:12:52 jongfoster
883 * Replacing function pointer in struct gateway with a directly
884 * called function forwarded_connect().
885 * Replacing struct gateway with struct forward_spec
887 * Revision 1.18 2001/06/03 19:12:16 oes
888 * introduced new cgi handling
890 * Revision 1.17 2001/06/01 20:07:23 jongfoster
891 * Now uses action +image-blocker{} rather than config->tinygif
893 * Revision 1.16 2001/06/01 18:49:17 jongfoster
894 * Replaced "list_share" with "list" - the tiny memory gain was not
895 * worth the extra complexity.
897 * Revision 1.15 2001/05/31 21:24:47 jongfoster
898 * Changed "permission" to "action" throughout.
899 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
900 * Moved vanilla wafer check into chat(), since we must now
901 * decide whether or not to add it based on the URL.
903 * Revision 1.14 2001/05/29 20:14:01 joergs
904 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
907 * Revision 1.13 2001/05/29 09:50:24 jongfoster
908 * Unified blocklist/imagelist/permissionslist.
909 * File format is still under discussion, but the internal changes
912 * Also modified interceptor behaviour:
913 * - We now intercept all URLs beginning with one of the following
914 * prefixes (and *only* these prefixes):
916 * * http://ijbswa.sf.net/config/
917 * * http://ijbswa.sourceforge.net/config/
918 * - New interceptors "home page" - go to http://i.j.b/ to see it.
919 * - Internal changes so that intercepted and fast redirect pages
920 * are not replaced with an image.
921 * - Interceptors now have the option to send a binary page direct
922 * to the client. (i.e. ijb-send-banner uses this)
923 * - Implemented show-url-info interceptor. (Which is why I needed
924 * the above interceptors changes - a typical URL is
925 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
926 * The previous mechanism would not have intercepted that, and
927 * if it had been intercepted then it then it would have replaced
930 * Revision 1.12 2001/05/27 22:17:04 oes
932 * - re_process_buffer no longer writes the modified buffer
933 * to the client, which was very ugly. It now returns the
934 * buffer, which it is then written by chat.
936 * - content_length now adjusts the Content-Length: header
937 * for modified documents rather than crunch()ing it.
938 * (Length info in csp->content_length, which is 0 for
939 * unmodified documents)
941 * - For this to work, sed() is called twice when filtering.
943 * Revision 1.11 2001/05/26 17:27:53 jongfoster
944 * Added support for CLF and fixed LOG_LEVEL_LOG.
945 * Also did CRLF->LF fix of my previous patch.
947 * Revision 1.10 2001/05/26 15:26:15 jongfoster
948 * ACL feature now provides more security by immediately dropping
949 * connections from untrusted hosts.
951 * Revision 1.9 2001/05/26 00:28:36 jongfoster
952 * Automatic reloading of config file.
953 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
954 * Most of the global variables have been moved to a new
955 * struct configuration_spec, accessed through csp->config->globalname
956 * Most of the globals remaining are used by the Win32 GUI.
958 * Revision 1.8 2001/05/25 22:43:18 jongfoster
959 * Fixing minor memory leak and buffer overflow.
961 * Revision 1.7 2001/05/25 22:34:30 jongfoster
964 * Revision 1.6 2001/05/23 00:13:58 joergs
965 * AmigaOS support fixed.
967 * Revision 1.5 2001/05/22 18:46:04 oes
969 * - Enabled filtering banners by size rather than URL
970 * by adding patterns that replace all standard banner
971 * sizes with the "Junkbuster" gif to the re_filterfile
973 * - Enabled filtering WebBugs by providing a pattern
974 * which kills all 1x1 images
976 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
977 * which is selected by the (nonstandard and therefore
978 * capital) letter 'U' in the option string.
979 * It causes the quantifiers to be ungreedy by default.
980 * Appending a ? turns back to greedy (!).
982 * - Added a new interceptor ijb-send-banner, which
983 * sends back the "Junkbuster" gif. Without imagelist or
984 * MSIE detection support, or if tinygif = 1, or the
985 * URL isn't recognized as an imageurl, a lame HTML
986 * explanation is sent instead.
988 * - Added new feature, which permits blocking remote
989 * script redirects and firing back a local redirect
991 * The feature is conditionally compiled, i.e. it
992 * can be disabled with --disable-fast-redirects,
993 * plus it must be activated by a "fast-redirects"
994 * line in the config file, has its own log level
995 * and of course wants to be displayed by show-proxy-args
996 * Note: Boy, all the #ifdefs in 1001 locations and
997 * all the fumbling with configure.in and acconfig.h
998 * were *way* more work than the feature itself :-(
1000 * - Because a generic redirect template was needed for
1001 * this, tinygif = 3 now uses the same.
1003 * - Moved GIFs, and other static HTTP response templates
1006 * - Some minor fixes
1008 * - Removed some >400 CRs again (Jon, you really worked
1011 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1012 * Made failure to bind() a fatal error.
1014 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1015 * Version 2.9.4 checkin.
1016 * - Merged popupfile and cookiefile, and added control over PCRS
1017 * filtering, in new "permissionsfile".
1018 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1019 * file error you now get a message box (in the Win32 GUI) rather
1020 * than the program exiting with no explanation.
1021 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1023 * - Removed tabs from "config"
1024 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1025 * - Bumped up version number.
1027 * Revision 1.2 2001/05/17 22:34:44 oes
1028 * - Added hint on GIF char array generation to jcc.c
1029 * - Cleaned CRLF's from the sources and related files
1030 * - Repaired logging for REF and FRC
1032 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1033 * Initial import of version 2.9.3 source tree
1036 *********************************************************************/
1042 #include <sys/types.h>
1051 # ifndef FEATURE_PTHREAD
1055 # include <windows.h>
1056 # include <process.h>
1057 # endif /* ndef FEATURE_PTHREAD */
1060 # ifndef _WIN_CONSOLE
1061 # include "w32log.h"
1062 # endif /* ndef _WIN_CONSOLE */
1063 # include "w32svrapi.h"
1065 #else /* ifndef _WIN32 */
1067 # if !defined (__OS2__)
1068 # include <unistd.h>
1069 # include <sys/wait.h>
1070 # endif /* ndef __OS2__ */
1071 # include <sys/time.h>
1072 # include <sys/stat.h>
1073 # include <sys/ioctl.h>
1076 #include <sys/termios.h>
1084 # include <signal.h>
1087 # include <socket.h> /* BeOS has select() for sockets only. */
1088 # include <OS.h> /* declarations for threads and stuff. */
1091 # if defined(__EMX__) || defined(__OS2__)
1092 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1097 #define bzero(B,N) memset(B,0x00,n)
1101 # include <select.h>
1106 #include "project.h"
1109 #include "filters.h"
1110 #include "loaders.h"
1111 #include "parsers.h"
1112 #include "miscutil.h"
1114 #include "jbsockets.h"
1115 #include "gateway.h"
1116 #include "actions.h"
1118 #include "loadcfg.h"
1119 #include "urlmatch.h"
1121 const char jcc_h_rcs[] = JCC_H_VERSION;
1122 const char project_h_rcs[] = PROJECT_H_VERSION;
1125 struct client_state clients[1];
1126 struct file_list files[1];
1128 #ifdef FEATURE_STATISTICS
1129 int urls_read = 0; /* total nr of urls read inc rejected */
1130 int urls_rejected = 0; /* total nr of urls rejected */
1131 #endif /* def FEATURE_STATISTICS */
1133 #ifdef FEATURE_GRACEFUL_TERMINATION
1134 int g_terminate = 0;
1137 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1138 static void sig_handler(int the_signal);
1140 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1141 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1142 static jb_err get_server_headers(struct client_state *csp);
1143 static const char *crunch_reason(const struct http_response *rsp);
1144 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1146 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1148 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1149 static jb_err change_request_destination(struct client_state *csp);
1150 static void chat(struct client_state *csp);
1151 static void serve(struct client_state *csp);
1152 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1153 static void usage(const char *myname);
1155 static void initialize_mutexes(void);
1156 static jb_socket bind_port_helper(struct configuration_spec *config);
1157 static void listen_loop(void);
1160 void serve(struct client_state *csp);
1161 #else /* ifndef AMIGA */
1162 static void serve(struct client_state *csp);
1163 #endif /* def AMIGA */
1166 static int32 server_thread(void *data);
1167 #endif /* def __BEOS__ */
1170 #define sleep(N) Sleep(((N) * 1000))
1174 #define sleep(N) DosSleep(((N) * 100))
1177 #ifdef MUTEX_LOCKS_AVAILABLE
1179 * XXX: Does the locking stuff really belong in this file?
1181 privoxy_mutex_t log_mutex;
1182 privoxy_mutex_t log_init_mutex;
1184 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1185 privoxy_mutex_t resolver_mutex;
1186 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1188 #ifndef HAVE_GMTIME_R
1189 privoxy_mutex_t gmtime_mutex;
1190 #endif /* ndef HAVE_GMTIME_R */
1192 #ifndef HAVE_LOCALTIME_R
1193 privoxy_mutex_t localtime_mutex;
1194 #endif /* ndef HAVE_GMTIME_R */
1197 privoxy_mutex_t rand_mutex;
1198 #endif /* ndef HAVE_RANDOM */
1200 #endif /* def MUTEX_LOCKS_AVAILABLE */
1203 const char *basedir = NULL;
1204 const char *pidfile = NULL;
1205 static int received_hup_signal = 0;
1206 #endif /* defined unix */
1208 /* HTTP snipplets. */
1209 static const char CSUCCEED[] =
1210 "HTTP/1.0 200 Connection established\r\n"
1211 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1213 static const char CHEADER[] =
1214 "HTTP/1.0 400 Invalid header received from client\r\n"
1215 "Proxy-Agent: Privoxy " VERSION "\r\n"
1216 "Content-Type: text/plain\r\n"
1217 "Connection: close\r\n\r\n"
1218 "Invalid header received from client.\r\n";
1220 static const char FTP_RESPONSE[] =
1221 "HTTP/1.0 400 Invalid request received from client\r\n"
1222 "Content-Type: text/plain\r\n"
1223 "Connection: close\r\n\r\n"
1224 "Invalid request. Privoxy doesn't support FTP.\r\n";
1226 static const char GOPHER_RESPONSE[] =
1227 "HTTP/1.0 400 Invalid request received from client\r\n"
1228 "Content-Type: text/plain\r\n"
1229 "Connection: close\r\n\r\n"
1230 "Invalid request. Privoxy doesn't support gopher.\r\n";
1232 /* XXX: should be a template */
1233 static const char MISSING_DESTINATION_RESPONSE[] =
1234 "HTTP/1.0 400 Bad request received from client\r\n"
1235 "Proxy-Agent: Privoxy " VERSION "\r\n"
1236 "Content-Type: text/plain\r\n"
1237 "Connection: close\r\n\r\n"
1238 "Bad request. Privoxy was unable to extract the destination.\r\n";
1240 /* XXX: should be a template */
1241 static const char NO_SERVER_DATA_RESPONSE[] =
1242 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1243 "Proxy-Agent: Privoxy " VERSION "\r\n"
1244 "Content-Type: text/plain\r\n"
1245 "Connection: close\r\n\r\n"
1246 "Empty server or forwarder response.\r\n"
1247 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1249 /* XXX: should be a template */
1250 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1251 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1252 "Proxy-Agent: Privoxy " VERSION "\r\n"
1253 "Content-Type: text/plain\r\n"
1254 "Connection: close\r\n\r\n"
1255 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1258 /* XXX: should be a template */
1259 static const char NULL_BYTE_RESPONSE[] =
1260 "HTTP/1.0 400 Bad request received from client\r\n"
1261 "Proxy-Agent: Privoxy " VERSION "\r\n"
1262 "Content-Type: text/plain\r\n"
1263 "Connection: close\r\n\r\n"
1264 "Bad request. Null byte(s) before end of request.\r\n";
1267 /* XXX: should be a template */
1268 static const char MESSED_UP_REQUEST_RESPONSE[] =
1269 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1270 "Proxy-Agent: Privoxy " VERSION "\r\n"
1271 "Content-Type: text/plain\r\n"
1272 "Connection: close\r\n\r\n"
1273 "Bad request. Messed up with header filters.\r\n";
1275 /* A function to crunch a response */
1276 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1278 /* Crunch function flags */
1279 #define CF_NO_FLAGS 0
1280 /* Cruncher applies to forced requests as well */
1281 #define CF_IGNORE_FORCE 1
1282 /* Crunched requests are counted for the block statistics */
1283 #define CF_COUNT_AS_REJECT 2
1285 /* A crunch function and its flags */
1288 const crunch_func_ptr cruncher;
1292 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1294 /* Complete list of cruncher functions */
1295 static const struct cruncher crunchers_all[] = {
1296 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1297 { block_url, CF_COUNT_AS_REJECT },
1298 #ifdef FEATURE_TRUST
1299 { trust_url, CF_COUNT_AS_REJECT },
1300 #endif /* def FEATURE_TRUST */
1301 { redirect_url, CF_NO_FLAGS },
1302 { dispatch_cgi, CF_IGNORE_FORCE},
1306 /* Light version, used after tags are applied */
1307 static const struct cruncher crunchers_light[] = {
1308 { block_url, CF_COUNT_AS_REJECT },
1309 { redirect_url, CF_NO_FLAGS },
1315 * XXX: Don't we really mean
1321 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1322 /*********************************************************************
1324 * Function : sig_handler
1326 * Description : Signal handler for different signals.
1327 * Exit gracefully on TERM and INT
1328 * or set a flag that will cause the errlog
1329 * to be reopened by the main thread on HUP.
1332 * 1 : the_signal = the signal cause this function to call
1336 *********************************************************************/
1337 static void sig_handler(int the_signal)
1343 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1355 received_hup_signal = 1;
1361 * We shouldn't be here, unless we catch signals
1362 * in main() that we can't handle here!
1364 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1372 /*********************************************************************
1374 * Function : client_protocol_is_unsupported
1376 * Description : Checks if the client used a known unsupported
1377 * protocol and deals with it by sending an error
1381 * 1 : csp = Current client state (buffers, headers, etc...)
1382 * 2 : req = the first request line send by the client
1384 * Returns : TRUE if an error response has been generated, or
1385 * FALSE if the request doesn't look invalid.
1387 *********************************************************************/
1388 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1391 * If it's a FTP or gopher request, we don't support it.
1393 * These checks are better than nothing, but they might
1394 * not work in all configurations and some clients might
1395 * have problems digesting the answer.
1397 * They should, however, never cause more problems than
1398 * Privoxy's old behaviour (returning the misleading HTML
1401 * "Could not resolve http://(ftp|gopher)://example.org").
1403 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1405 const char *response = NULL;
1406 const char *protocol = NULL;
1408 if (!strncmpic(req, "GET ftp://", 10))
1410 response = FTP_RESPONSE;
1415 response = GOPHER_RESPONSE;
1416 protocol = "GOPHER";
1418 log_error(LOG_LEVEL_ERROR,
1419 "%s tried to use Privoxy as %s proxy: %s",
1420 csp->ip_addr_str, protocol, req);
1421 log_error(LOG_LEVEL_CLF,
1422 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1424 write_socket(csp->cfd, response, strlen(response));
1433 /*********************************************************************
1435 * Function : get_request_destination_elsewhere
1437 * Description : If the client's request was redirected into
1438 * Privoxy without the client's knowledge,
1439 * the request line lacks the destination host.
1441 * This function tries to get it elsewhere,
1442 * provided accept-intercepted-requests is enabled.
1444 * "Elsewhere" currently only means "Host: header",
1445 * but in the future we may ask the redirecting
1446 * packet filter to look the destination up.
1448 * If the destination stays unknown, an error
1449 * response is send to the client and headers
1450 * are freed so that chat() can return directly.
1453 * 1 : csp = Current client state (buffers, headers, etc...)
1454 * 2 : headers = a header list
1456 * Returns : JB_ERR_OK if the destination is now known, or
1457 * JB_ERR_PARSE if it isn't.
1459 *********************************************************************/
1460 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1464 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1466 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1467 " Privoxy isn't configured to accept intercepted requests.",
1468 csp->ip_addr_str, csp->http->cmd);
1469 /* XXX: Use correct size */
1470 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1471 csp->ip_addr_str, csp->http->cmd);
1473 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1474 destroy_list(headers);
1476 return JB_ERR_PARSE;
1478 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1480 /* Split the domain we just got for pattern matching */
1481 init_domain_components(csp->http);
1487 /* We can't work without destination. Go spread the news.*/
1489 req = list_to_text(headers);
1491 /* XXX: Use correct size */
1492 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1493 csp->ip_addr_str, csp->http->cmd);
1494 log_error(LOG_LEVEL_ERROR,
1495 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1496 csp->ip_addr_str, csp->http->cmd, req);
1499 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1500 destroy_list(headers);
1502 return JB_ERR_PARSE;
1505 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1506 * to get the destination IP address, use it as host directly
1507 * or do a reverse DNS lookup first.
1512 /*********************************************************************
1514 * Function : get_server_headers
1516 * Description : Parses server headers in iob and fills them
1517 * into csp->headers so that they can later be
1521 * 1 : csp = Current client state (buffers, headers, etc...)
1523 * Returns : JB_ERR_OK if everything went fine, or
1524 * JB_ERR_PARSE if the headers were incomplete.
1526 *********************************************************************/
1527 static jb_err get_server_headers(struct client_state *csp)
1529 int continue_hack_in_da_house = 0;
1532 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1537 * continue hack in da house. Ignore the ending of
1538 * this head and continue enlisting header lines.
1539 * The reason is described below.
1541 enlist(csp->headers, "");
1542 continue_hack_in_da_house = 0;
1545 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1548 * It's a bodyless continue response, don't
1549 * stop header parsing after reaching its end.
1551 * As a result Privoxy will concatenate the
1552 * next response's head and parse and deliver
1553 * the headers as if they belonged to one request.
1555 * The client will separate them because of the
1556 * empty line between them.
1558 * XXX: What we're doing here is clearly against
1559 * the intended purpose of the continue header,
1560 * and under some conditions (HTTP/1.0 client request)
1561 * it's a standard violation.
1563 * Anyway, "sort of against the spec" is preferable
1564 * to "always getting confused by Continue responses"
1565 * (Privoxy's behaviour before this hack was added)
1567 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1568 continue_hack_in_da_house = 1;
1570 else if (*header == '\0')
1573 * If the header is empty, but the Continue hack
1574 * isn't active, we can assume that we reached the
1575 * end of the buffer before we hit the end of the
1578 * Inform the caller an let it decide how to handle it.
1580 return JB_ERR_PARSE;
1583 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1586 * XXX: Should we quit the request and return a
1587 * out of memory error page instead?
1589 log_error(LOG_LEVEL_ERROR,
1590 "Out of memory while enlisting server headers. %s lost.",
1600 /*********************************************************************
1602 * Function : crunch_reason
1604 * Description : Translates the crunch reason code into a string.
1607 * 1 : rsp = a http_response
1609 * Returns : A string with the crunch reason or an error description.
1611 *********************************************************************/
1612 static const char *crunch_reason(const struct http_response *rsp)
1614 char * reason = NULL;
1616 assert(rsp != NULL);
1619 return "Internal error while searching for crunch reason";
1622 switch (rsp->reason)
1624 case RSP_REASON_UNSUPPORTED:
1625 reason = "Unsupported HTTP feature";
1627 case RSP_REASON_BLOCKED:
1630 case RSP_REASON_UNTRUSTED:
1631 reason = "Untrusted";
1633 case RSP_REASON_REDIRECTED:
1634 reason = "Redirected";
1636 case RSP_REASON_CGI_CALL:
1637 reason = "CGI Call";
1639 case RSP_REASON_NO_SUCH_DOMAIN:
1640 reason = "DNS failure";
1642 case RSP_REASON_FORWARDING_FAILED:
1643 reason = "Forwarding failed";
1645 case RSP_REASON_CONNECT_FAILED:
1646 reason = "Connection failure";
1648 case RSP_REASON_OUT_OF_MEMORY:
1649 reason = "Out of memory (may mask other reasons)";
1652 reason = "No reason recorded";
1660 /*********************************************************************
1662 * Function : send_crunch_response
1664 * Description : Delivers already prepared response for
1665 * intercepted requests, logs the interception
1666 * and frees the response.
1669 * 1 : csp = Current client state (buffers, headers, etc...)
1670 * 1 : rsp = Fully prepared response. Will be freed on exit.
1672 * Returns : Nothing.
1674 *********************************************************************/
1675 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1677 const struct http_request *http = csp->http;
1678 char status_code[4];
1680 assert(rsp != NULL);
1681 assert(rsp->head != NULL);
1686 * Not supposed to happen. If it does
1687 * anyway, treat it as an unknown error.
1689 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1690 /* return code doesn't matter */
1695 /* If rsp is still NULL, we have serious internal problems. */
1696 log_error(LOG_LEVEL_FATAL,
1697 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1701 * Extract the status code from the actual head
1702 * that was send to the client. It is the only
1703 * way to get it right for all requests, including
1704 * the fixed ones for out-of-memory problems.
1706 * A head starts like this: 'HTTP/1.1 200...'
1710 status_code[0] = rsp->head[9];
1711 status_code[1] = rsp->head[10];
1712 status_code[2] = rsp->head[11];
1713 status_code[3] = '\0';
1715 /* Write the answer to the client */
1716 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1717 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1719 /* There is nothing we can do about it. */
1720 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1723 /* Log that the request was crunched and why. */
1724 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1725 http->hostport, http->path, crunch_reason(rsp));
1726 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1727 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1729 /* Clean up and return */
1730 if (cgi_error_memory() != rsp)
1732 free_http_response(rsp);
1739 /*********************************************************************
1741 * Function : request_contains_null_bytes
1743 * Description : Checks for NULL bytes in the request and sends
1744 * an error message to the client if any were found.
1746 * XXX: currently not used, see comment in chat().
1749 * 1 : csp = Current client state (buffers, headers, etc...)
1750 * 2 : buf = Data from the client's request to check.
1751 * 3 : len = The data length.
1753 * Returns : TRUE if the request contained one or more NULL bytes, or
1756 *********************************************************************/
1757 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1759 size_t c_len; /* Request lenght when treated as C string */
1761 c_len = strlen(buf);
1766 * Null byte(s) found. Log the request,
1767 * return an error response and hang up.
1769 size_t tmp_len = c_len;
1774 * Replace NULL byte(s) with '°' characters
1775 * so the request can be logged as string.
1776 * XXX: Is there a better replacement character?
1779 tmp_len += strlen(buf+tmp_len);
1780 } while (tmp_len < len);
1782 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1783 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1784 log_error(LOG_LEVEL_HEADER,
1785 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1787 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1789 /* XXX: Log correct size */
1790 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1800 /*********************************************************************
1802 * Function : crunch_response_triggered
1804 * Description : Checks if the request has to be crunched,
1805 * and delivers the crunch response if necessary.
1808 * 1 : csp = Current client state (buffers, headers, etc...)
1809 * 2 : crunchers = list of cruncher functions to run
1811 * Returns : TRUE if the request was answered with a crunch response
1814 *********************************************************************/
1815 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1817 struct http_response *rsp = NULL;
1818 const struct cruncher *c;
1821 * If CGI request crunching is disabled,
1822 * check the CGI dispatcher out of order to
1823 * prevent unintentional blocks or redirects.
1825 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1826 && (NULL != (rsp = dispatch_cgi(csp))))
1828 /* Deliver, log and free the interception response. */
1829 send_crunch_response(csp, rsp);
1833 for (c = crunchers; c->cruncher != NULL; c++)
1836 * Check the cruncher if either Privoxy is toggled
1837 * on and the request isn't forced, or if the cruncher
1838 * applies to forced requests as well.
1840 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1841 !(csp->flags & CSP_FLAG_FORCED)) ||
1842 (c->flags & CF_IGNORE_FORCE))
1844 rsp = c->cruncher(csp);
1847 /* Deliver, log and free the interception response. */
1848 send_crunch_response(csp, rsp);
1849 #ifdef FEATURE_STATISTICS
1850 if (c->flags & CF_COUNT_AS_REJECT)
1852 csp->flags |= CSP_FLAG_REJECTED;
1854 #endif /* def FEATURE_STATISTICS */
1865 /*********************************************************************
1867 * Function : build_request_line
1869 * Description : Builds the HTTP request line.
1871 * If a HTTP forwarder is used it expects the whole URL,
1872 * web servers only get the path.
1875 * 1 : csp = Current client state (buffers, headers, etc...)
1876 * 2 : fwd = The forwarding spec used for the request
1877 * XXX: Should use http->fwd instead.
1878 * 3 : request_line = The old request line which will be replaced.
1880 * Returns : Nothing. Terminates in case of memory problems.
1882 *********************************************************************/
1883 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1885 struct http_request *http = csp->http;
1887 assert(http->ssl == 0);
1890 * Downgrade http version from 1.1 to 1.0
1891 * if +downgrade action applies.
1893 if ( (csp->action->flags & ACTION_DOWNGRADE)
1894 && (!strcmpic(http->ver, "HTTP/1.1")))
1897 http->ver = strdup("HTTP/1.0");
1899 if (http->ver == NULL)
1901 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1906 * Rebuild the request line.
1908 freez(*request_line);
1909 *request_line = strdup(http->gpc);
1910 string_append(request_line, " ");
1912 if (fwd->forward_host)
1914 string_append(request_line, http->url);
1918 string_append(request_line, http->path);
1920 string_append(request_line, " ");
1921 string_append(request_line, http->ver);
1923 if (*request_line == NULL)
1925 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1927 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1931 /*********************************************************************
1933 * Function : change_request_destination
1935 * Description : Parse a (rewritten) request line and regenerate
1936 * the http request data.
1939 * 1 : csp = Current client state (buffers, headers, etc...)
1941 * Returns : Forwards the parse_http_request() return code.
1942 * Terminates in case of memory problems.
1944 *********************************************************************/
1945 static jb_err change_request_destination(struct client_state *csp)
1947 struct http_request *http = csp->http;
1950 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1951 free_http_request(http);
1952 err = parse_http_request(csp->headers->first->str, http, csp);
1953 if (JB_ERR_OK != err)
1955 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1956 jb_err_to_string(err));
1960 /* XXX: ocmd is a misleading name */
1961 http->ocmd = strdup(http->cmd);
1962 if (http->ocmd == NULL)
1964 log_error(LOG_LEVEL_FATAL,
1965 "Out of memory copying rewritten HTTP request line");
1973 /*********************************************************************
1977 * Description : Once a connection to the client has been accepted,
1978 * this function is called (via serve()) to handle the
1979 * main business of the communication. When this
1980 * function returns, the caller must close the client
1983 * FIXME: chat is nearly thousand lines long.
1987 * 1 : csp = Current client state (buffers, headers, etc...)
1989 * Returns : Nothing.
1991 *********************************************************************/
1992 static void chat(struct client_state *csp)
1994 char buf[BUFFER_SIZE];
2002 int ms_iis5_hack = 0;
2003 size_t byte_count = 0;
2004 int forwarded_connect_retries = 0;
2005 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2006 const struct forward_spec * fwd;
2007 struct http_request *http;
2008 int len; /* for buffer sizes (and negative error codes) */
2011 /* Function that does the content filtering for the current request */
2012 filter_function_ptr content_filter = NULL;
2014 /* Skeleton for HTTP response, if we should intercept the request */
2015 struct http_response *rsp;
2017 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2018 struct list header_list;
2019 struct list *headers = &header_list;
2023 memset(buf, 0, sizeof(buf));
2026 * Read the client's request. Note that since we're not using select() we
2027 * could get blocked here if a client connected, then didn't say anything!
2032 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2034 if (len <= 0) break; /* error! */
2037 * If there is no memory left for buffering the
2038 * request, there is nothing we can do but hang up
2040 if (add_to_iob(csp, buf, len))
2045 req = get_header(csp->iob);
2047 } while ((NULL != req) && ('\0' == *req));
2049 if ((NULL != req) && ('\0' != *req))
2051 /* Request received. Validate and parse it. */
2055 * XXX: Temporary disabled to prevent problems
2056 * with POST requests whose bodies are allowed to
2057 * contain NULL bytes. BR#1730105.
2059 * The main purpose of this check is to properly
2060 * log stuff like BitTorrent traffic and other junk
2061 * that hits public proxies. It's not required for
2062 * Privoxy to functions as those requests are discarded
2065 * It probably should be rewritten to only check
2066 * the head of the request. Another option would
2067 * be to let all POST requests pass, although that
2068 * may not be good enough.
2070 if (request_contains_null_bytes(csp, buf, len))
2072 /* NULL bytes found and dealt with, just hang up. */
2077 /* Does the request line look invalid? */
2078 if (client_protocol_is_unsupported(csp, req))
2081 * Yes. The request has already been
2082 * answered with a error response, the buffers
2083 * were freed and we're done with chatting.
2088 #ifdef FEATURE_FORCE_LOAD
2090 * If this request contains the FORCE_PREFIX and blocks
2091 * aren't enforced, get rid of it and set the force flag.
2093 if (strstr(req, FORCE_PREFIX))
2095 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2097 log_error(LOG_LEVEL_FORCE,
2098 "Ignored force prefix in request: \"%s\".", req);
2102 strclean(req, FORCE_PREFIX);
2103 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2104 csp->flags |= CSP_FLAG_FORCED;
2108 #endif /* def FEATURE_FORCE_LOAD */
2109 err = parse_http_request(req, http, csp);
2110 if (JB_ERR_OK != err)
2112 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2118 if (http->cmd == NULL)
2120 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2121 /* XXX: Use correct size */
2122 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2123 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2125 free_http_request(http);
2129 /* grab the rest of the client's headers */
2133 p = get_header(csp->iob);
2137 /* There are no additional headers to read. */
2144 * We didn't receive a complete header
2145 * line yet, get the rest of it.
2147 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2150 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2151 destroy_list(headers);
2155 if (add_to_iob(csp, buf, len))
2158 * If there is no memory left for buffering the
2159 * request, there is nothing we can do but hang up
2161 destroy_list(headers);
2168 * We were able to read a complete
2169 * header and can finaly enlist it.
2176 if (http->host == NULL)
2179 * If we still don't know the request destination,
2180 * the request is invalid or the client uses
2181 * Privoxy without its knowledge.
2183 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2186 * Our attempts to get the request destination
2187 * elsewhere failed or Privoxy is configured
2188 * to only accept proxy requests.
2190 * An error response has already been send
2191 * and we're done here.
2198 * Determine the actions for this URL
2200 #ifdef FEATURE_TOGGLE
2201 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2203 /* Most compatible set of actions (i.e. none) */
2204 init_current_action(csp->action);
2207 #endif /* ndef FEATURE_TOGGLE */
2209 get_url_actions(csp, http);
2213 * Save a copy of the original request for logging
2215 http->ocmd = strdup(http->cmd);
2217 if (http->ocmd == NULL)
2219 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2222 enlist(csp->headers, http->cmd);
2224 /* Append the previously read headers */
2225 list_append_list_unique(csp->headers, headers);
2226 destroy_list(headers);
2228 err = sed(csp, FILTER_CLIENT_HEADERS);
2229 if (JB_ERR_OK != err)
2231 assert(err == JB_ERR_PARSE);
2232 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2234 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2236 /* Check request line for rewrites. */
2237 if ((NULL == csp->headers->first->str)
2238 || (strcmp(http->cmd, csp->headers->first->str) &&
2239 (JB_ERR_OK != change_request_destination(csp))))
2242 * A header filter broke the request line - bail out.
2244 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2245 /* XXX: Use correct size */
2246 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2247 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2249 free_http_request(http);
2253 /* decide how to route the HTTP request */
2254 fwd = forward_url(csp, http);
2257 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2258 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2262 * build the http request to send to the server
2263 * we have to do one of the following:
2265 * create = use the original HTTP request to create a new
2266 * HTTP request that has either the path component
2267 * without the http://domainspec (w/path) or the
2268 * full orininal URL (w/url)
2269 * Note that the path and/or the HTTP version may
2270 * have been altered by now.
2272 * connect = Open a socket to the host:port of the server
2273 * and short-circuit server and client socket.
2275 * pass = Pass the request unchanged if forwarding a CONNECT
2276 * request to a parent proxy. Note that we'll be sending
2277 * the CFAIL message ourselves if connecting to the parent
2278 * fails, but we won't send a CSUCCEED message if it works,
2279 * since that would result in a double message (ours and the
2280 * parent's). After sending the request to the parent, we simply
2283 * here's the matrix:
2286 * +--------+--------+
2288 * 0 | create | connect|
2290 * Forwarding +--------+--------+
2292 * 1 | create | pass |
2294 * +--------+--------+
2298 if (http->ssl && connect_port_is_forbidden(csp))
2300 const char *acceptable_connect_ports =
2301 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2302 assert(NULL != acceptable_connect_ports);
2303 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2304 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2305 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2306 csp->action->flags |= ACTION_BLOCK;
2312 freez(csp->headers->first->str);
2313 build_request_line(csp, fwd, &csp->headers->first->str);
2316 hdr = list_to_text(csp->headers);
2319 /* FIXME Should handle error properly */
2320 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2324 * We have a request. Check if one of the crunchers wants it.
2326 if (crunch_response_triggered(csp, crunchers_all))
2329 * Yes. The client got the crunch response
2330 * and we are done here after cleaning up.
2333 list_remove_all(csp->headers);
2339 * The headers can't be removed earlier because
2340 * they were still needed for the referrer check
2341 * in case of CGI crunches.
2343 * XXX: Would it be worth to move the referrer check
2344 * into client_referrer() and set a flag if it's trusted?
2346 list_remove_all(csp->headers);
2348 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2350 if (fwd->forward_host)
2352 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2353 fwd->forward_host, fwd->forward_port, http->hostport);
2357 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2360 /* here we connect to the server, gateway, or the forwarder */
2362 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2363 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2365 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2366 forwarded_connect_retries, http->hostport);
2369 if (csp->sfd == JB_INVALID_SOCKET)
2371 if (fwd->type != SOCKS_NONE)
2374 rsp = error_response(csp, "forwarding-failed", errno);
2376 else if (errno == EINVAL)
2378 rsp = error_response(csp, "no-such-domain", errno);
2382 rsp = error_response(csp, "connect-failed", errno);
2383 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2388 /* Write the answer to the client */
2391 send_crunch_response(csp, rsp);
2398 if (fwd->forward_host || (http->ssl == 0))
2400 /* write the client's (modified) header to the server
2401 * (along with anything else that may be in the buffer)
2404 if (write_socket(csp->sfd, hdr, strlen(hdr))
2405 || (flush_socket(csp->sfd, csp->iob) < 0))
2407 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2410 rsp = error_response(csp, "connect-failed", errno);
2414 send_crunch_response(csp, rsp);
2424 * We're running an SSL tunnel and we're not forwarding,
2425 * so just send the "connect succeeded" message to the
2426 * client, flush the rest, and get out of the way.
2428 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2436 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2438 /* we're finished with the client's header */
2441 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2443 /* pass data between the client and server
2444 * until one or the other shuts down the connection.
2453 * FD_ZERO here seems to point to an errant macro which crashes.
2454 * So do this by hand for now...
2456 memset(&rfds,0x00,sizeof(fd_set));
2460 FD_SET(csp->cfd, &rfds);
2461 FD_SET(csp->sfd, &rfds);
2463 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2467 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2471 /* this is the body of the browser's request
2472 * just read it and write it.
2475 if (FD_ISSET(csp->cfd, &rfds))
2477 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2481 break; /* "game over, man" */
2484 if (write_socket(csp->sfd, buf, (size_t)len))
2486 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2493 * The server wants to talk. It could be the header or the body.
2494 * If `hdr' is null, then it's the header otherwise it's the body.
2495 * FIXME: Does `hdr' really mean `host'? No.
2499 if (FD_ISSET(csp->sfd, &rfds))
2502 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2506 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2508 if (http->ssl && (fwd->forward_host == NULL))
2511 * Just hang up. We already confirmed the client's CONNECT
2512 * request with status code 200 and unencrypted content is
2513 * no longer welcome.
2515 log_error(LOG_LEVEL_ERROR,
2516 "CONNECT already confirmed. Unable to tell the client about the problem.");
2519 else if (byte_count)
2522 * Just hang up. We already transmitted the original headers
2523 * and parts of the original content and therefore missed the
2524 * chance to send an error message (without risking data corruption).
2526 * XXX: we could retry with a fancy range request here.
2528 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2529 "Unable to tell the client about the problem.");
2533 rsp = error_response(csp, "connect-failed", errno);
2537 send_crunch_response(csp, rsp);
2543 /* Add a trailing zero. This lets filter_popups
2544 * use string operations.
2548 /* Normally, this would indicate that we've read
2549 * as much as the server has sent us and we can
2550 * close the client connection. However, Microsoft
2551 * in its wisdom has released IIS/5 with a bug that
2552 * prevents it from sending the trailing \r\n in
2553 * a 302 redirect header (and possibly other headers).
2554 * To work around this if we've haven't parsed
2555 * a full header we'll append a trailing \r\n
2556 * and see if this now generates a valid one.
2558 * This hack shouldn't have any impacts. If we've
2559 * already transmitted the header or if this is a
2560 * SSL connection, then we won't bother with this
2561 * hack. So we only work on partially received
2562 * headers. If we append a \r\n and this still
2563 * doesn't generate a valid header, then we won't
2564 * transmit anything to the client.
2569 if (server_body || http->ssl)
2572 * If we have been buffering up the document,
2573 * now is the time to apply content modification
2574 * and send the result to the client.
2578 p = execute_content_filter(csp, content_filter);
2580 * If the content filter fails, use the original
2581 * buffer and length.
2582 * (see p != NULL ? p : csp->iob->cur below)
2586 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2589 if (JB_ERR_OK != update_server_headers(csp))
2591 log_error(LOG_LEVEL_FATAL,
2592 "Failed to update server headers. after filtering.");
2595 hdr = list_to_text(csp->headers);
2598 /* FIXME Should handle error properly */
2599 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2602 if (write_socket(csp->cfd, hdr, strlen(hdr))
2603 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2605 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2615 break; /* "game over, man" */
2619 * This is NOT the body, so
2620 * Let's pretend the server just sent us a blank line.
2622 snprintf(buf, sizeof(buf), "\r\n");
2623 len = (int)strlen(buf);
2626 * Now, let the normal header parsing algorithm below do its
2627 * job. If it fails, we'll exit instead of continuing.
2634 * If this is an SSL connection or we're in the body
2635 * of the server document, just write it to the client,
2636 * unless we need to buffer the body for later content-filtering
2639 if (server_body || http->ssl)
2644 * If there is no memory left for buffering the content, or the buffer limit
2645 * has been reached, switch to non-filtering mode, i.e. make & write the
2646 * header, flush the iob and buf, and get out of the way.
2648 if (add_to_iob(csp, buf, len))
2653 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2655 hdr = list_to_text(csp->headers);
2659 * Memory is too tight to even generate the header.
2660 * Send our static "Out-of-memory" page.
2662 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2663 rsp = cgi_error_memory();
2664 send_crunch_response(csp, rsp);
2668 hdrlen = strlen(hdr);
2670 if (write_socket(csp->cfd, hdr, hdrlen)
2671 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2672 || (write_socket(csp->cfd, buf, (size_t)len)))
2674 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2681 * Reset the byte_count to the amount of bytes
2682 * we just flushed. len will be added a few lines below,
2683 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2685 byte_count = (size_t)flushed;
2687 content_filter = NULL;
2693 if (write_socket(csp->cfd, buf, (size_t)len))
2695 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2699 byte_count += (size_t)len;
2704 /* we're still looking for the end of the
2705 * server's header ... (does that make header
2706 * parsing an "out of body experience" ?
2710 * buffer up the data we just read. If that fails,
2711 * there's little we can do but send our static
2712 * out-of-memory page.
2714 if (add_to_iob(csp, buf, len))
2716 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2717 rsp = cgi_error_memory();
2718 send_crunch_response(csp, rsp);
2723 /* Convert iob into something sed() can digest */
2724 if (JB_ERR_PARSE == get_server_headers(csp))
2728 /* Well, we tried our MS IIS/5
2729 * hack and it didn't work.
2730 * The header is incomplete
2731 * and there isn't anything
2732 * we can do about it.
2734 log_error(LOG_LEVEL_INFO,
2735 "MS IIS5 hack didn't produce valid headers.");
2740 /* Since we have to wait for
2741 * more from the server before
2742 * we can parse the headers
2743 * we just continue here.
2749 /* Did we actually get anything? */
2750 if (NULL == csp->headers->first)
2752 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2753 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2754 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2755 free_http_request(http);
2759 assert(csp->headers->first->str);
2761 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2762 strncmpic(csp->headers->first->str, "ICY", 3))
2765 * It doesn't look like a HTTP (or Shoutcast) response:
2766 * tell the client and log the problem.
2768 if (strlen(csp->headers->first->str) > 30)
2770 csp->headers->first->str[30] = '\0';
2772 log_error(LOG_LEVEL_ERROR,
2773 "Invalid server or forwarder response. Starts with: %s",
2774 csp->headers->first->str);
2775 log_error(LOG_LEVEL_CLF,
2776 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2777 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2778 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2779 free_http_request(http);
2784 * We have now received the entire server header,
2785 * filter it and send the result to the client
2787 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2789 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2791 hdr = list_to_text(csp->headers);
2794 /* FIXME Should handle error properly */
2795 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2798 if (crunch_response_triggered(csp, crunchers_light))
2801 * One of the tags created by a server-header
2802 * tagger triggered a crunch. We already
2803 * delivered the crunch response to the client
2804 * and are done here after cleaning up.
2809 /* Buffer and pcrs filter this if appropriate. */
2811 if (!http->ssl) /* We talk plaintext */
2813 content_filter = get_filter_function(csp);
2816 * Only write if we're not buffering for content modification
2818 if (!content_filter)
2820 /* write the server's (modified) header to
2821 * the client (along with anything else that
2822 * may be in the buffer)
2825 if (write_socket(csp->cfd, hdr, strlen(hdr))
2826 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2828 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2830 /* the write failed, so don't bother
2831 * mentioning it to the client...
2832 * it probably can't hear us anyway.
2838 byte_count += (size_t)len;
2841 /* we're finished with the server's header */
2846 /* If this was a MS IIS/5 hack then it means
2847 * the server has already closed the
2848 * connection. Nothing more to read. Time
2853 log_error(LOG_LEVEL_INFO,
2854 "Closed server connection detected with MS IIS5 hack enabled.");
2861 return; /* huh? we should never get here */
2864 if (csp->content_length == 0)
2867 * If Privoxy didn't recalculate the
2868 * Content-Lenght, byte_count is still
2871 csp->content_length = byte_count;
2874 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2875 csp->ip_addr_str, http->ocmd, csp->content_length);
2879 /*********************************************************************
2883 * Description : This is little more than chat. We only "serve" to
2884 * to close any socket that chat may have opened.
2887 * 1 : csp = Current client state (buffers, headers, etc...)
2891 *********************************************************************/
2893 void serve(struct client_state *csp)
2894 #else /* ifndef AMIGA */
2895 static void serve(struct client_state *csp)
2896 #endif /* def AMIGA */
2899 close_socket(csp->cfd);
2901 if (csp->sfd != JB_INVALID_SOCKET)
2903 close_socket(csp->sfd);
2906 csp->flags &= ~CSP_FLAG_ACTIVE;
2912 /*********************************************************************
2914 * Function : server_thread
2916 * Description : We only exist to call `serve' in a threaded environment.
2919 * 1 : data = Current client state (buffers, headers, etc...)
2921 * Returns : Always 0.
2923 *********************************************************************/
2924 static int32 server_thread(void *data)
2926 serve((struct client_state *) data);
2933 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2934 /*********************************************************************
2938 * Description : Print usage info & exit.
2940 * Parameters : Pointer to argv[0] for identifying ourselves
2944 *********************************************************************/
2945 static void usage(const char *myname)
2947 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2951 #endif /* defined(unix) */
2954 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2955 #endif /* defined(unix) */
2956 "[--version] [configfile]\n"
2957 "Aborting\n", myname);
2962 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2965 #ifdef MUTEX_LOCKS_AVAILABLE
2966 /*********************************************************************
2968 * Function : privoxy_mutex_lock
2970 * Description : Locks a mutex.
2973 * 1 : mutex = The mutex to lock.
2975 * Returns : Void. May exit in case of errors.
2977 *********************************************************************/
2978 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
2980 #ifdef FEATURE_PTHREAD
2981 int err = pthread_mutex_lock(mutex);
2984 if (mutex != &log_mutex)
2986 log_error(LOG_LEVEL_FATAL,
2987 "Mutex locking failed: %s.\n", strerror(err));
2992 EnterCriticalSection(mutex);
2993 #endif /* def FEATURE_PTHREAD */
2997 /*********************************************************************
2999 * Function : privoxy_mutex_unlock
3001 * Description : Unlocks a mutex.
3004 * 1 : mutex = The mutex to unlock.
3006 * Returns : Void. May exit in case of errors.
3008 *********************************************************************/
3009 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3011 #ifdef FEATURE_PTHREAD
3012 int err = pthread_mutex_unlock(mutex);
3015 if (mutex != &log_mutex)
3017 log_error(LOG_LEVEL_FATAL,
3018 "Mutex unlocking failed: %s.\n", strerror(err));
3023 LeaveCriticalSection(mutex);
3024 #endif /* def FEATURE_PTHREAD */
3028 /*********************************************************************
3030 * Function : privoxy_mutex_init
3032 * Description : Prepares a mutex.
3035 * 1 : mutex = The mutex to initialize.
3037 * Returns : Void. May exit in case of errors.
3039 *********************************************************************/
3040 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3042 #ifdef FEATURE_PTHREAD
3043 int err = pthread_mutex_init(mutex, 0);
3046 printf("Fatal error. Mutex initialization failed: %s.\n",
3051 InitializeCriticalSection(mutex);
3052 #endif /* def FEATURE_PTHREAD */
3054 #endif /* def MUTEX_LOCKS_AVAILABLE */
3056 /*********************************************************************
3058 * Function : initialize_mutexes
3060 * Description : Prepares mutexes if mutex support is available.
3064 * Returns : Void, exits in case of errors.
3066 *********************************************************************/
3067 static void initialize_mutexes(void)
3069 #ifdef MUTEX_LOCKS_AVAILABLE
3071 * Prepare global mutex semaphores
3073 privoxy_mutex_init(&log_mutex);
3075 privoxy_mutex_init(&log_init_mutex);
3078 * XXX: The assumptions below are a bit naive
3079 * and can cause locks that aren't necessary.
3081 * For example older FreeBSD versions (< 6.x?)
3082 * have no gethostbyname_r, but gethostbyname is
3085 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3086 privoxy_mutex_init(&resolver_mutex);
3087 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3089 * XXX: should we use a single mutex for
3090 * localtime() and gmtime() as well?
3092 #ifndef HAVE_GMTIME_R
3093 privoxy_mutex_init(&gmtime_mutex);
3094 #endif /* ndef HAVE_GMTIME_R */
3096 #ifndef HAVE_LOCALTIME_R
3097 privoxy_mutex_init(&localtime_mutex);
3098 #endif /* ndef HAVE_GMTIME_R */
3101 privoxy_mutex_init(&rand_mutex);
3102 #endif /* ndef HAVE_RANDOM */
3103 #endif /* def MUTEX_LOCKS_AVAILABLE */
3107 /*********************************************************************
3111 * Description : Load the config file and start the listen loop.
3112 * This function is a lot more *sane* with the `load_config'
3113 * and `listen_loop' functions; although it stills does
3114 * a *little* too much for my taste.
3117 * 1 : argc = Number of parameters (including $0).
3118 * 2 : argv = Array of (char *)'s to the parameters.
3120 * Returns : 1 if : can't open config file, unrecognized directive,
3121 * stats requested in multi-thread mode, can't open the
3122 * log file, can't open the jar file, listen port is invalid,
3123 * any load fails, and can't bind port.
3125 * Else main never returns, the process must be signaled
3126 * to terminate execution. Or, on Windows, use the
3127 * "File", "Exit" menu option.
3129 *********************************************************************/
3131 int real_main(int argc, const char *argv[])
3133 int main(int argc, const char *argv[])
3137 unsigned int random_seed;
3139 struct passwd *pw = NULL;
3140 struct group *grp = NULL;
3143 char *pre_chroot_nslookup_to_load_resolver = NULL;
3150 #if !defined(_WIN32)
3158 * Parse the command line arguments
3160 * XXX: simply printing usage information in case of
3161 * invalid arguments isn't particularly user friendly.
3163 while (++argc_pos < argc)
3166 /* Check to see if the service must be installed or uninstalled */
3167 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3169 const char *pName = argv[argc_pos] + 9;
3172 exit( (install_service(pName)) ? 0 : 1 );
3174 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3176 const char *pName = argv[argc_pos] + 11;
3179 exit((uninstall_service(pName)) ? 0 : 1);
3181 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3183 bRunAsService = TRUE;
3184 w32_set_service_cwd();
3185 atexit(w32_service_exit_notify);
3188 #endif /* defined(_WIN32) */
3191 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3193 if (strcmp(argv[argc_pos], "--help") == 0)
3198 else if(strcmp(argv[argc_pos], "--version") == 0)
3200 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3206 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3211 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3213 if (++argc_pos == argc) usage(argv[0]);
3214 pidfile = strdup(argv[argc_pos]);
3217 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3219 if (++argc_pos == argc) usage(argv[argc_pos]);
3221 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3224 if (NULL == (grp = getgrnam(p)))
3226 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3230 if (NULL == (pw = getpwnam(argv[argc_pos])))
3232 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3235 if (p != NULL) *--p = '\0';
3238 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3240 if (++argc_pos == argc) usage(argv[0]);
3241 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3244 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3248 #endif /* defined(unix) */
3250 else if (argc_pos + 1 != argc)
3253 * This is neither the last command line
3254 * option, nor was it recognized before,
3255 * therefore it must be invalid.
3261 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3263 configfile = argv[argc_pos];
3266 } /* -END- while (more arguments) */
3269 if ( *configfile != '/' )
3271 char cwd[BUFFER_SIZE];
3273 size_t abs_file_size;
3275 /* make config-filename absolute here */
3276 if (NULL == getcwd(cwd, sizeof(cwd)))
3278 perror("failed to get current working directory");
3283 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3284 basedir = strdup(cwd);
3286 if (NULL == basedir ||
3287 NULL == (abs_file = malloc(abs_file_size)))
3289 perror("malloc failed");
3292 strlcpy(abs_file, basedir, abs_file_size);
3293 strlcat(abs_file, "/", abs_file_size );
3294 strlcat(abs_file, configfile, abs_file_size);
3295 configfile = abs_file;
3297 #endif /* defined unix */
3301 clients->next = NULL;
3303 /* XXX: factor out initialising after the next stable release. */
3306 #elif defined(_WIN32)
3310 /* Prepare mutexes if supported and necessary. */
3311 initialize_mutexes();
3313 /* Enable logging until further notice. */
3314 init_log_module(Argv[0]);
3316 random_seed = (unsigned int)time(NULL);
3318 srandom(random_seed);
3321 #endif /* ifdef HAVE_RANDOM */
3324 * Unix signal handling
3326 * Catch the abort, interrupt and terminate signals for a graceful exit
3327 * Catch the hangup signal so the errlog can be reopened.
3328 * Ignore the broken pipe signals (FIXME: Why?)
3330 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3333 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3334 const int ignored_signals[] = { SIGPIPE, 0 };
3336 for (idx = 0; catched_signals[idx] != 0; idx++)
3338 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3339 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3341 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3342 #endif /* ifdef sun */
3344 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3348 for (idx = 0; ignored_signals[idx] != 0; idx++)
3350 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3352 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3357 #else /* ifdef _WIN32 */
3358 # ifdef _WIN_CONSOLE
3360 * We *are* in a windows console app.
3361 * Print a verbose messages about FAQ's and such
3363 printf("%s", win32_blurb);
3364 # endif /* def _WIN_CONSOLE */
3365 #endif /* def _WIN32 */
3368 /* Initialize the CGI subsystem */
3369 cgi_init_error_messages();
3372 * If runnig on unix and without the --nodaemon
3373 * option, become a daemon. I.e. fork, detach
3374 * from tty and get process group leadership
3387 if ( pid < 0 ) /* error */
3392 else if ( pid != 0 ) /* parent */
3397 * must check for errors
3398 * child died due to missing files aso
3401 wpid = waitpid( pid, &status, WNOHANG );
3410 /* Should be more portable, but not as well tested */
3415 #else /* ndef __FreeBSD__ */
3417 #endif /* ndef __FreeBSD__ */
3418 fd = open("/dev/tty", O_RDONLY);
3421 /* no error check here */
3422 ioctl( fd, TIOCNOTTY,0 );
3427 * stderr (fd 2) will be closed later on, when the
3428 * log file has been parsed.
3435 } /* -END- if (!no_daemon) */
3438 * As soon as we have written the PID file, we can switch
3439 * to the user and group ID indicated by the --user option
3445 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3447 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3453 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3455 /* Read the time zone file from /etc before doing chroot. */
3457 if (NULL != pre_chroot_nslookup_to_load_resolver
3458 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3460 /* Initialize resolver library. */
3461 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3463 if (chroot(pw->pw_dir) < 0)
3465 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3469 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3472 if (setuid(pw->pw_uid))
3474 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3478 char putenv_dummy[64];
3480 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3481 if (putenv(putenv_dummy) != 0)
3483 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3486 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3487 if (putenv(putenv_dummy) != 0)
3489 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3495 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3498 #endif /* defined unix */
3501 /* This will be FALSE unless the command line specified --service
3505 /* Yup, so now we must attempt to establish a connection
3506 * with the service dispatcher. This will only work if this
3507 * process was launched by the service control manager to
3508 * actually run as a service. If this isn't the case, i've
3509 * known it take around 30 seconds or so for the call to return.
3512 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3513 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3515 /* Service has run, and at this point is now being stopped, so just return */
3520 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3522 /* An error occurred. Usually it's because --service was wrongly specified
3523 * and we were unable to connect to the Service Control Dispatcher because
3524 * it wasn't expecting us and is therefore not listening.
3526 * For now, just continue below to call the listen_loop function.
3529 #endif /* def _WIN32 */
3539 /*********************************************************************
3541 * Function : bind_port_helper
3543 * Description : Bind the listen port. Handles logging, and aborts
3547 * 1 : config = Privoxy configuration. Specifies port
3550 * Returns : Port that was opened.
3552 *********************************************************************/
3553 static jb_socket bind_port_helper(struct configuration_spec * config)
3558 if (config->haddr == NULL)
3560 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3565 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3566 config->hport, config->haddr);
3569 result = bind_port(config->haddr, config->hport, &bfd);
3576 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3577 "There may be another Privoxy or some other "
3578 "proxy running on port %d",
3579 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3580 config->hport, config->hport);
3583 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3584 "The hostname is not resolvable",
3585 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3588 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3589 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3592 /* shouldn't get here */
3593 return JB_INVALID_SOCKET;
3596 config->need_bind = 0;
3603 /* Without this simple workaround we get this compiler warning from _beginthread
3604 * warning C4028: formal parameter 1 different from declaration
3606 void w32_service_listen_loop(void *p)
3610 #endif /* def _WIN32 */
3613 /*********************************************************************
3615 * Function : listen_loop
3617 * Description : bind the listen port and enter a "FOREVER" listening loop.
3623 *********************************************************************/
3624 static void listen_loop(void)
3626 struct client_state *csp = NULL;
3628 struct configuration_spec * config;
3630 config = load_config();
3632 bfd = bind_port_helper(config);
3634 #ifdef FEATURE_GRACEFUL_TERMINATION
3635 while (!g_terminate)
3640 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3641 while (waitpid(-1, NULL, WNOHANG) > 0)
3643 /* zombie children */
3645 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3648 * Free data that was used by died threads
3654 * Re-open the errlog after HUP signal
3656 if (received_hup_signal)
3658 init_error_log(Argv[0], config->logfile);
3659 received_hup_signal = 0;
3663 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3665 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3669 csp->flags |= CSP_FLAG_ACTIVE;
3670 csp->sfd = JB_INVALID_SOCKET;
3672 csp->config = config = load_config();
3674 if ( config->need_bind )
3677 * Since we were listening to the "old port", we will not see
3678 * a "listen" param change until the next IJB request. So, at
3679 * least 1 more request must be made for us to find the new
3680 * setting. I am simply closing the old socket and binding the
3683 * Which-ever is correct, we will serve 1 more page via the
3684 * old settings. This should probably be a "show-proxy-args"
3685 * request. This should not be a so common of an operation
3686 * that this will hurt people's feelings.
3691 bfd = bind_port_helper(config);
3694 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3696 if (!accept_connection(csp, bfd))
3698 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3711 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3714 #ifdef FEATURE_TOGGLE
3715 if (global_toggle_state)
3716 #endif /* def FEATURE_TOGGLE */
3718 csp->flags |= CSP_FLAG_TOGGLED_ON;
3721 if (run_loader(csp))
3723 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3724 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3728 if (block_acl(NULL,csp))
3730 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3731 close_socket(csp->cfd);
3735 #endif /* def FEATURE_ACL */
3737 /* add it to the list of clients */
3738 csp->next = clients->next;
3739 clients->next = csp;
3741 if (config->multi_threaded)
3745 /* this is a switch () statment in the C preprocessor - ugh */
3746 #undef SELECTED_ONE_OPTION
3748 /* Use Pthreads in preference to native code */
3749 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3750 #define SELECTED_ONE_OPTION
3752 pthread_t the_thread;
3753 pthread_attr_t attrs;
3755 pthread_attr_init(&attrs);
3756 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3757 errno = pthread_create(&the_thread, &attrs,
3758 (void * (*)(void *))serve, csp);
3759 child_id = errno ? -1 : 0;
3760 pthread_attr_destroy(&attrs);
3764 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3765 #define SELECTED_ONE_OPTION
3766 child_id = _beginthread(
3767 (void (*)(void *))serve,
3772 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3773 #define SELECTED_ONE_OPTION
3774 child_id = _beginthread(
3775 (void(* _Optlink)(void*))serve,
3781 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3782 #define SELECTED_ONE_OPTION
3784 thread_id tid = spawn_thread
3785 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3787 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3789 child_id = (int) tid;
3798 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3799 #define SELECTED_ONE_OPTION
3800 csp->cfd = ReleaseSocket(csp->cfd, -1);
3803 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3804 NP_Output, Output(),
3805 NP_CloseOutput, FALSE,
3806 NP_Name, (ULONG)"privoxy child",
3810 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3811 NP_Output, Output(),
3812 NP_CloseOutput, FALSE,
3813 NP_Name, (ULONG)"privoxy child",
3814 NP_StackSize, 200*1024,
3820 ((struct Task *)child_id)->tc_UserData = csp;
3821 Signal((struct Task *)child_id, SIGF_SINGLE);
3826 #if !defined(SELECTED_ONE_OPTION)
3829 /* This block is only needed when using fork().
3830 * When using threads, the server thread was
3831 * created and run by the call to _beginthread().
3833 if (child_id == 0) /* child */
3836 #ifdef FEATURE_TOGGLE
3837 int inherited_toggle_state = global_toggle_state;
3838 #endif /* def FEATURE_TOGGLE */
3843 * If we've been toggled or we've blocked the request, tell Mom
3846 #ifdef FEATURE_TOGGLE
3847 if (inherited_toggle_state != global_toggle_state)
3849 rc |= RC_FLAG_TOGGLED;
3851 #endif /* def FEATURE_TOGGLE */
3853 #ifdef FEATURE_STATISTICS
3854 if (csp->flags & CSP_FLAG_REJECTED)
3856 rc |= RC_FLAG_BLOCKED;
3858 #endif /* ndef FEATURE_STATISTICS */
3862 else if (child_id > 0) /* parent */
3864 /* in a fork()'d environment, the parent's
3865 * copy of the client socket and the CSP
3869 #if !defined(_WIN32) && !defined(__CYGWIN__)
3871 wait( &child_status );
3874 * Evaluate child's return code: If the child has
3875 * - been toggled, toggle ourselves
3876 * - blocked its request, bump up the stats counter
3879 #ifdef FEATURE_TOGGLE
3880 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3882 global_toggle_state = !global_toggle_state;
3884 #endif /* def FEATURE_TOGGLE */
3886 #ifdef FEATURE_STATISTICS
3888 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3892 #endif /* def FEATURE_STATISTICS */
3894 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3895 close_socket(csp->cfd);
3896 csp->flags &= ~CSP_FLAG_ACTIVE;
3900 #undef SELECTED_ONE_OPTION
3901 /* end of cpp switch () */
3903 if (child_id < 0) /* failed */
3905 char buf[BUFFER_SIZE];
3907 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3909 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3911 write_socket(csp->cfd, buf, strlen(buf));
3912 close_socket(csp->cfd);
3913 csp->flags &= ~CSP_FLAG_ACTIVE;
3924 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3926 /* Clean up. Aim: free all memory (no leaks) */
3927 #ifdef FEATURE_GRACEFUL_TERMINATION
3929 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3931 unload_current_config_file();
3932 unload_current_actions_file();
3933 unload_current_re_filterfile();
3934 #ifdef FEATURE_TRUST
3935 unload_current_trust_file();
3938 if (config->multi_threaded)
3945 } while ((clients->next != NULL) && (--i > 0));
3949 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3960 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3961 /* Cleanup - remove taskbar icon etc. */
3966 #endif /* FEATURE_GRACEFUL_TERMINATION */
projects / privoxy.git / blob