1 const char jcc_rcs[] = "$Id: jcc.c,v 1.178 2008/05/10 13:23:38 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 1.178 2008/05/10 13:23:38 fabiankeil
37 * Don't provide get_header() with the whole client state
38 * structure when it only needs access to csp->iob.
40 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
41 * Make the "read the rest of the headers" loop a bit more readable.
43 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
44 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
45 * - Remove useless comment.
47 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
48 * Fix comment grammar.
50 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
51 * Remove the pointless buffer in client_protocol_is_unsupported().
53 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
54 * Least-effort fix for bug #1821930 (reported by Lee):
55 * If the response doesn't look like HTTP,
56 * tell the client and log the problem.
58 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
59 * Don't pass the whole csp structure to flush_socket()
60 * when it only needs a file descriptor and a buffer.
62 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
63 * Remove kill-popups action.
65 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
66 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
68 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
69 * Remove the treat-forbidden-connects-like-blocks action. We now
70 * use the "blocked" page for forbidden CONNECT requests by default.
72 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
73 * Also use shiny new connect_port_is_forbidden() in jcc.c.
75 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
76 * Rename url_actions() to get_url_actions() and let it
77 * use the standard parameter ordering.
79 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
80 * Let forward_url() use the standard parameter ordering
81 * and mark its second parameter immutable.
83 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
84 * Remove the "Listening ... for local connections only" log message.
85 * Whether or not remote connections are able to reach Privoxy is up
86 * to the operating system.
88 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
89 * Prevent the log messages for CONNECT requests to unacceptable
90 * ports from printing the limit-connect argument as [null] if
91 * limit-connect hasn't been explicitly enabled.
93 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
94 * Make sure all console-mode apps get a usage() instance
96 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
97 * Reword NO_SERVER_DATA_RESPONSE to make it harder
98 * to misunderstand what the message is all about.
100 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
101 * Unbreak trustfile which previously didn't work without
102 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
104 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
105 * Plug memory leak. Spotted by Valgrind, triggered by
106 * Privoxy-Regression-Test feeding proxyfuzz.py.
108 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
109 * In the HTTP snipplets, refer to the client as client.
111 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
112 * Emit a log message when activating the MS IIS5 hack.
114 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
115 * Log the "weak randomization factor" warning only
116 * once for mingw32 and provide some more details.
118 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
119 * Initialize log module after initializing mutexes, future
120 * deadlocks in that code should now work cross-platform.
122 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
123 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
124 * Reported by Bert van Leeuwen in BR#1818808.
126 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
127 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
129 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
130 * When in daemon mode, close stderr after the configuration file has been
131 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
133 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
134 * - Fix a crash when parsing invalid requests whose first header
135 * is rejected by get_header(). Regression (re?)introduced
136 * in r1.143 by yours truly.
137 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
138 * client_cookie_adder() to make sure send-vanilla-wafer can be
139 * controlled through tags (and thus regression-tested).
141 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
142 * - Move get_filter_function() from jcc.c to filters.c
143 * so the filter functions can be static.
144 * - Don't bother filtering body-less responses.
146 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
147 * Execute content filters through execute_content_filter().
149 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
150 * Initialize req to NULL to make sure it's defined if the
151 * first read_socket() call fails. Reported by icmp30.
153 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
154 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
155 * extensive comments moved to user manual.
157 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
158 * Don't crash if a broken header filter wiped out the request line.
160 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
161 * Fix byte_count calculation in case of flushes
162 * and don't parse the server headers a second time.
164 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
165 * - If there's a connection problem after we already forwarded
166 * parts of the original content, just hang up. Fixes BR#1776724.
167 * - Fix warnings about unused code on mingw32.
168 * - In case of flushes, calculate the byte count
169 * less incorrectly (I think).
171 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
172 * Add some more prototypes for static functions.
174 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
175 * Comment out request_contains_null_bytes() until it's used again.
177 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
178 * #1763173 from Stefan Huehner: s@const static@static const@
179 * and declare some more functions static.
181 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
182 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
183 * and explain why they were rejected in the first place.
184 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
185 * CONNECT requests. The request line was missing.
186 * - Add two more XXX reminders as we don't have enough already.
188 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
189 * As Hal noticed, checking dispatch_cgi() as the last cruncher
190 * looks like a bug if CGI requests are blocked unintentionally,
191 * so don't do it unless the user enabled the new config option
192 * "allow-cgi-request-crunching".
194 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
195 * - Allow to rewrite the request destination behind the client's back.
196 * - Turn the weird-looking unconditional for loop that
197 * reads the client request into a conditional while loop.
198 * Move the stuff that only runs once out of the loop.
199 * - Move parts of chat(), server_content_type() and the
200 * necessary stuff to fix BR#1750917 into get_filter_function().
202 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
203 * Temporary workaround for BR#1730105.
205 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
206 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
207 * deadlocks and crashes on OpenBSD and possibly other OS with neither
208 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
209 * Thanks to Ralf Horstmann for report and solution.
211 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
212 * Add forward-override{} to change the forwarding settings through
213 * action sections. This is mainly interesting to forward different
214 * clients differently (for example based on User-Agent or request
217 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
218 * - Let usage() mention the --chroot parameter.
219 * - Use read_socket() consistently and always leave
220 * the last buffer byte alone, even in cases where
221 * null termination (currently) doesn't matter.
223 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
224 * - Fix config file loading on Unix if no config file is specified.
225 * Since r1.97 Privoxy would always interpret the last argument as
226 * config file, even if it's a valid command line option.
227 * - Abort in case of unrecognized command line options. Closes #1719696.
228 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
229 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
231 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
232 * - Don't rerun crunchers that only depend on the request URL.
233 * - Don't count redirects and CGI requests as "blocked requests".
235 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
236 * Support crunching based on tags created by server-header taggers.
238 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
239 * Make HTTP snippets static (again). Add a Content-Type for those
240 * with content so the browser doesn't guess it based on the URL.
242 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
243 * Move crunching and request line rebuilding out of chat().
245 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
246 * Introduce tags as alternative way to specify which
247 * actions apply to a request. At the moment tags can be
248 * created based on client and server headers.
250 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
251 * Don't CLF-log CONNECT requests twice.
253 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
254 * Log the source address for ACL-related connection drops.
256 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
257 * New config option: enforce-blocks.
259 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
260 * - Move null byte check into separate function.
261 * - Don't confuse the client with error pages
262 * if a CONNECT request was already confirmed.
264 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
265 * Speed up NULL byte escaping and only log the complete
266 * NULL byte requests with header debugging enabled.
268 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
269 * Answer requests that contain NULL bytes with
270 * a custom response instead of waiting for more
271 * data until the client eventually hangs up.
273 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
274 * - Move delivery and logging of crunched responses
275 * from chat() into send_crunch_response().
276 * - Display the reason for generating http_responses.
277 * - Log the content length for LOG_LEVEL_CLF correctly
278 * (still incorrect for some fixed responses).
279 * - Reword an incorrect comment about
280 * treat-forbidden-connects-like-blocks violating
282 * - Add some log messages.
284 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
285 * Move mutex initialization into separate
286 * function and exit in case of errors.
288 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
289 * - Start to reduce chat()'s line count and move
290 * parts of it into separate functions.
291 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
293 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
294 * - Add Proxy-Agent header to HTTP snippets that are
295 * supposed to reach HTTP clients only.
296 * - Made a few CONNECT log messages more descriptive.
297 * - Catch completely empty server responses (as seen
298 * with Tor's fake ".noconnect" top level domain).
299 * - Use shiny new "forwarding-failed" template for socks errors.
301 * Revision 1.118 2007/01/07 07:43:43 joergs
302 * AmigaOS4 support added.
304 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
305 * Added config option accept-intercepted-requests
306 * and disabled it by default.
308 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
309 * Reverted parts of my last commit
310 * to keep error handling working.
312 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
313 * Fixed gcc43 conversion warnings.
315 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
316 * Fix -pedantic ISO C warning about converting
317 * from function pointer to object pointer.
319 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
320 * Silence compiler warning I introduced with my last commit.
322 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
323 * Mutex protect rand() if POSIX threading
324 * is used, warn the user if that's not possible
325 * and stop using it on _WIN32 where it could
328 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
329 * Don't prevent core dumps by catching SIGABRT.
330 * It's rude and makes debugging unreasonable painful.
332 * Revision 1.110 2006/12/13 14:52:53 etresoft
333 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
335 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
336 * Privoxy is now able to run as intercepting
337 * proxy in combination with any packet filter
338 * that does the port redirection. The destination
339 * is extracted from the "Host:" header which
340 * should be available for nearly all requests.
342 * Moved HTTP snipplets into jcc.c.
343 * Added error message for gopher proxy requests.
345 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
346 * Only unlink the pidfile if it's actually used.
348 * Change order of interception checks to make
349 * it possible to block or redirect requests for
352 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
353 * Make pthread mutex locking more generic. Instead of
354 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
355 * and use mutex locking unless there is an _r function
356 * available. Better safe than sorry.
358 * Fixes "./configure --disable-pthread" and should result
359 * in less threading-related problems on pthread-using platforms,
360 * but it still doesn't fix BR#1122404.
362 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
363 * Move pthread.h inclusion from jcc.c to jcc.h.
364 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
366 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
367 * Don't exit after receiving the second SIGHUP on Solaris.
369 * Fixes BR 1052235, but the same problem may exist on other
370 * systems. Once 3.0.6 is out we should use sigset()
371 * where available and see if it breaks anything.
373 * Revision 1.104 2006/09/23 13:26:38 roro
374 * Replace TABs by spaces in source code.
376 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
377 * Fix +redirect{}. Didn't work with -fast-redirects.
379 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
380 * Respond with 400 and a short text message
381 * if the client tries to use Privoxy as FTP proxy.
383 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
384 * Make number of retries in case of forwarded-connect problems
385 * a config file option (forwarded-connect-retries) and use 0 as
388 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
389 * Set random(3) seed.
391 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
392 * Follow the OpenBSD port's lead and protect the resolve
393 * functions on OpenBSD as well.
395 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
396 * --user fix. Only use the user as group if no group is specified.
397 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
399 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
400 * Windows service (re-)integration
404 * --install[:service_name]
405 * --uninstall[:service_name]
408 * They work as follows:
409 * --install will create a service for you and then terminate.
410 * By default the service name will be "privoxy" (without the quotes).
411 * However you can run multiple services if you wish, just by adding
412 * a colon and then a name (no spaces).
414 * --uninstall follows the exact same rules a --install.
416 * --service is used when the program is executed by the service
417 * control manager, and in normal circumstances would never be
418 * used as a command line argument.
420 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
421 * Windows service integration
423 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
424 * Incorporate Fabian Keil's patch work:
425 http://www.fabiankeil.de/sourcecode/privoxy/
427 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
428 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
429 * with what was really the latest development (the v_3_0_branch branch)
431 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
432 * Thanks to Jindrich Makovicka for a race condition fix for the log
433 * file. The race condition remains for non-pthread implementations.
434 * Reference patch #1175720.
436 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
437 * Add the ability to check jpeg images for invalid
438 * lengths of comment blocks. Defensive strategy
439 * against the exploit:
440 * Microsoft Security Bulletin MS04-028
441 * Buffer Overrun in JPEG Processing (GDI+) Could
442 * Allow Code Execution (833987)
443 * Enabled with +inspect-jpegs in actions files.
445 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
446 * - Fixed usage info for non-unix platforms
447 * - Fixed small cmdline parsing bug
449 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
450 * Diagnostics: Now preserve the returncode of pthread_create
451 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
453 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
454 * No longer ignore SIGCHLD. Fixes bug #769381
456 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
457 * Close jarfile on graceful exit, remove stray line
459 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
460 * Cosmetics: Killed a warning, a typo and an allocation left at exit
462 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
463 * No longer rely on non-POSIX.1 extensions of getcwd().
466 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
467 * Replaced setenv() by posix-compliant putenv()
468 * Thanks to Neil McCalden (nmcc AT users.sf.net).
470 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
471 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
473 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
474 * Clean-up and extension of improvements for forked mode:
475 * - Child's return code now consists of flags RC_FLAG_*
476 * - Reporting toggle to parent now properly #ifdef'ed
477 * - Children now report blocking to parent. This enables
478 * statistics in forked mode
480 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
481 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
482 * Children now report having been toggled through _exit(17), parents
483 * watch for that code and toggle themselves if found.
485 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
486 * Wrapping all *_r functions (the non-_r versions of them) with
487 * mutex semaphores for OSX. Hopefully this will take care of all
488 * of those pesky crash reports.
490 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
491 * Fixed two mostly harmless mem leaks
493 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
494 * Fix: Head of global clients list now initialized to NULL
496 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
497 * Added basic support for OPTIONS and TRACE HTTP methods:
498 * - New interceptor direct_response() added in chat().
499 * - sed() moved to earlier in the process, so that the
500 * Host: header is evaluated before actions and forwarding
503 * Revision 1.92 2002/05/08 16:00:46 oes
504 * Chat's buffer handling:
505 * - Fixed bug with unchecked out-of-mem conditions
506 * while reading client request & server headers
507 * - No longer predict if the buffer limit will be exceeded
508 * in the next read -- check add_to_iob's new
509 * return code. If buffer couldn't be extended
510 * (policy or out-of-mem) while
511 * - reading from client: abort
512 * - reading server headers: send error page
513 * - buffering server body for filter: flush,
514 * and if that fails: send error page
516 * Revision 1.91 2002/04/08 20:35:58 swa
519 * Revision 1.90 2002/04/02 14:57:28 oes
520 * Made sending wafers independent of FEATURE_COOKIE_JAR
522 * Revision 1.89 2002/03/31 17:18:59 jongfoster
523 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
525 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
526 * More compiler warning message maintenance
528 * Revision 1.87 2002/03/26 22:29:54 swa
529 * we have a new homepage!
531 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
532 * Workaround for closing the jarfile before load_config() comes around again
534 * Revision 1.85 2002/03/24 15:23:33 jongfoster
537 * Revision 1.84 2002/03/24 13:25:43 swa
538 * name change related issues
540 * Revision 1.83 2002/03/16 23:54:06 jongfoster
541 * Adding graceful termination feature, to help look for memory leaks.
542 * If you enable this (which, by design, has to be done by hand
543 * editing config.h) and then go to http://i.j.b/die, then the program
544 * will exit cleanly after the *next* request. It should free all the
545 * memory that was used.
547 * Revision 1.82 2002/03/13 00:27:05 jongfoster
550 * Revision 1.81 2002/03/12 01:42:50 oes
551 * Introduced modular filters
553 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
554 * OS/2 port maintenance:
555 * - Fixed EMX build - it had decayed a little
556 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
557 * substituted a memset for now.
559 * Revision 1.79 2002/03/09 20:03:52 jongfoster
560 * - Making various functions return int rather than size_t.
561 * (Undoing a recent change). Since size_t is unsigned on
562 * Windows, functions like read_socket that return -1 on
563 * error cannot return a size_t.
565 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
566 * crashes, and also frequently caused JB to jump to 100%
567 * CPU and stay there. (Because it thought it had just
568 * read ((unsigned)-1) == 4Gb of data...)
570 * - The signature of write_socket has changed, it now simply
571 * returns success=0/failure=nonzero.
573 * - Trying to get rid of a few warnings --with-debug on
574 * Windows, I've introduced a new type "jb_socket". This is
575 * used for the socket file descriptors. On Windows, this
576 * is SOCKET (a typedef for unsigned). Everywhere else, it's
577 * an int. The error value can't be -1 any more, so it's
578 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
579 * Windows it maps to the #define INVALID_SOCKET.)
581 * - The signature of bind_port has changed.
583 * Revision 1.78 2002/03/08 21:35:04 oes
584 * Added optional group supplement to --user option. Will now use default group of user if no group given
586 * Revision 1.77 2002/03/07 03:52:06 oes
587 * - Fixed compiler warnings etc
588 * - Improved handling of failed DNS lookups
590 * Revision 1.76 2002/03/06 22:54:35 jongfoster
591 * Automated function-comment nitpicking.
593 * Revision 1.75 2002/03/06 10:02:19 oes
594 * Fixed stupid bug when --user was not given
596 * Revision 1.74 2002/03/06 00:49:31 jongfoster
597 * Fixing warning on Windows
598 * Making #ifdefs that refer to the same variable consistently
599 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
601 * Revision 1.73 2002/03/05 23:57:30 hal9
602 * Stray character 's' on line 1618 was breaking build.
604 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
605 * - Re-enable OS/2 building after new parms were added
606 * - Fix false out of memory report when resolving CGI templates when no IP
607 * address is available of failed attempt (a la no such domain)
609 * Revision 1.71 2002/03/05 18:13:56 oes
610 * Added --user option
612 * Revision 1.70 2002/03/05 04:52:42 oes
613 * Deleted non-errlog debugging code
615 * Revision 1.69 2002/03/04 23:50:00 jongfoster
616 * Splitting off bind_port() call into bind_port_helper(), with
619 * Revision 1.68 2002/03/04 20:17:32 oes
622 * Revision 1.67 2002/03/04 18:18:57 oes
623 * - Removed _DEBUG mode
624 * - Cleand up cmdline parsing
625 * - Introduced --no-daemon, --pidfile options
626 * - Cleaned up signal handling:
627 * - Terminate cleanly on INT, TERM and ABRT
628 * - Schedule logfile for re-opening on HUP
629 * - Ignore CHLD and PIPE
630 * - Leave the rest with their default handlers
631 * - Uniform handler registration
632 * - Added usage() function
633 * - Played styleguide police
635 * Revision 1.66 2002/03/03 15:06:55 oes
636 * Re-enabled automatic config reloading
638 * Revision 1.65 2002/03/03 14:49:11 oes
639 * Fixed CLF logging: Now uses client's original HTTP request
641 * Revision 1.64 2002/03/03 09:18:03 joergs
642 * Made jumbjuster work on AmigaOS again.
644 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
645 * Clean up a little CRLF unpleasantness that suddenly appeared
647 * Revision 1.62 2002/02/20 23:17:23 jongfoster
648 * Detecting some out-of memory conditions and exiting with a log message.
650 * Revision 1.61 2002/01/17 21:01:52 jongfoster
651 * Moving all our URL and URL pattern parsing code to urlmatch.c.
653 * Revision 1.60 2001/12/30 14:07:32 steudten
654 * - Add signal handling (unix)
655 * - Add SIGHUP handler (unix)
656 * - Add creation of pidfile (unix)
657 * - Add action 'top' in rc file (RH)
658 * - Add entry 'SIGNALS' to manpage
659 * - Add exit message to logfile (unix)
661 * Revision 1.59 2001/12/13 14:07:18 oes
662 * Fixed Bug: 503 error page now sent OK
664 * Revision 1.58 2001/11/30 23:37:24 jongfoster
665 * Renaming the Win32 config file to config.txt - this is almost the
666 * same as the corresponding UNIX name "config"
668 * Revision 1.57 2001/11/16 00:47:43 jongfoster
669 * Changing the tty-disconnection code to use setsid().
671 * Revision 1.56 2001/11/13 20:20:54 jongfoster
672 * Tabs->spaces, fixing a bug with missing {} around an if()
674 * Revision 1.55 2001/11/13 20:14:53 jongfoster
675 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
677 * Revision 1.54 2001/11/07 00:03:14 steudten
678 * Give reliable return value if an error
679 * occurs not just 0 with new daemon mode.
681 * Revision 1.53 2001/11/05 21:41:43 steudten
682 * Add changes to be a real daemon just for unix os.
683 * (change cwd to /, detach from controlling tty, set
684 * process group and session leader to the own process.
686 * Add some fatal-error log message for failed malloc().
687 * Add '-d' if compiled with 'configure --with-debug' to
688 * enable debug output.
690 * Revision 1.52 2001/10/26 20:11:20 jongfoster
691 * Fixing type mismatch
693 * Revision 1.51 2001/10/26 17:38:28 oes
696 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
697 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
698 * threads to call select() simultaneously. So, it's time to do a real, live,
699 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
700 * (native). Both versions will work, but using __OS2__ offers multi-threading.
702 * Revision 1.49 2001/10/23 21:41:35 jongfoster
703 * Added call to initialize the (statically-allocated of course)
704 * "out of memory" CGI response.
706 * Revision 1.48 2001/10/10 19:56:46 jongfoster
707 * Moving some code that wasn't cookie-related out of an #ifdef
710 * Revision 1.47 2001/10/10 16:44:36 oes
711 * Added CONNECT destination port limitation check
713 * Revision 1.46 2001/10/08 15:17:41 oes
714 * Re-enabled SSL forwarding
716 * Revision 1.45 2001/10/07 15:42:11 oes
717 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
719 * Moved downgrading of the HTTP version from parse_http_request to
720 * chat(), since we can't decide if it is necessary before we have
721 * determined the actions for the URL. The HTTP command is now
722 * *always* re-built so the repairs need no longer be special-cased.
724 * filter_popups now gets a csp pointer so it can raise the new
725 * CSP_FLAG_MODIFIED flag.
729 * Added configurable size limit for the IOB. If the IOB grows so
730 * large that the next read would exceed the limit, the header
731 * is generated, and the header & unfiltered buffer are flushed
732 * to the client. Chat then continues in non-buffering,
733 * non-filtering body mode.
735 * Revision 1.44 2001/10/02 18:13:57 oes
738 * Revision 1.43 2001/10/02 15:32:13 oes
739 * Moved generation of hdr
741 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
742 * Cleaning up 2 compiler warnings on OS/2.
744 * Revision 1.41 2001/09/16 17:05:14 jongfoster
745 * Removing unused #include showarg.h
747 * Revision 1.40 2001/09/16 15:41:45 jongfoster
748 * Fixing signed/unsigned comparison warning.
750 * Revision 1.39 2001/09/16 13:21:27 jongfoster
751 * Changes to use new list functions.
753 * Revision 1.38 2001/09/16 13:01:46 jongfoster
754 * Removing redundant function call that zeroed zalloc()'d memory.
756 * Revision 1.37 2001/09/10 11:12:24 oes
757 * Deleted unused variable
759 * Revision 1.36 2001/09/10 10:56:15 oes
760 * Silenced compiler warnings
762 * Revision 1.35 2001/07/31 14:44:22 oes
763 * Deleted unused size parameter from filter_popups()
765 * Revision 1.34 2001/07/30 22:08:36 jongfoster
766 * Tidying up #defines:
767 * - All feature #defines are now of the form FEATURE_xxx
768 * - Permanently turned off WIN_GUI_EDIT
769 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
771 * Revision 1.33 2001/07/29 19:32:00 jongfoster
772 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
774 * Revision 1.32 2001/07/29 18:47:05 jongfoster
775 * Adding missing #include "loadcfg.h"
777 * Revision 1.31 2001/07/29 12:17:48 oes
778 * Applied pthread fix by Paul Lieverse
780 * Revision 1.30 2001/07/25 22:57:13 jongfoster
781 * __BEOS__ no longer overrides FEATURE_PTHREAD.
782 * This is because FEATURE_PTHREAD will soon be widely used, so I
783 * want to keep it simple.
785 * Revision 1.29 2001/07/24 12:47:06 oes
786 * Applied BeOS support update by Eugenia
788 * Revision 1.28 2001/07/23 13:26:12 oes
789 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
791 * Revision 1.27 2001/07/19 19:09:47 haroon
792 * - Added code to take care of the situation where while processing the first
793 * server response (which includes the server header), after finding the end
794 * of the headers we were not looking past the end of the headers for
795 * content modification. I enabled it for filter_popups.
796 * Someone else should look to see if other similar operations should be
797 * done to the discarded portion of the buffer.
799 * Note 2001/07/20: No, the other content modification mechanisms will process
800 * the whole iob later anyway. --oes
802 * Revision 1.26 2001/07/18 12:31:36 oes
805 * Revision 1.25 2001/07/15 19:43:49 jongfoster
806 * Supports POSIX threads.
807 * Also removed some unused #includes.
809 * Revision 1.24 2001/07/13 14:00:40 oes
810 * - Generic content modification scheme:
811 * Each feature has its own applicability flag that is set
812 * from csp->action->flags.
813 * Replaced the "filtering" int flag , by a function pointer
814 * "content_filter" to the function that will do the content
815 * modification. If it is != NULL, the document will be buffered
816 * and processed through *content_filter, which must set
817 * csp->content_length and return a modified copy of the body
818 * or return NULL (on failiure).
819 * - Changed csp->is_text to the more generic bitmap csp->content_type
820 * which can currently take the valued CT_TEXT or CT_GIF
822 * - Removed all #ifdef PCRS
824 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
825 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
826 * the `block_acl' call. This prevents a compilation error when the user
827 * does not wish to use the "ACL" feature.
829 * Revision 1.22 2001/06/29 21:45:41 oes
830 * Indentation, CRLF->LF, Tab-> Space
832 * Revision 1.21 2001/06/29 13:29:36 oes
833 * - Cleaned up, improved comments
834 * - Unified all possible interceptors (CGI,
835 * block, trust, fast_redirect) in one
836 * place, with one (CGI) answer generation
837 * mechansim. Much clearer now.
838 * - Removed the GIF image generation, which
839 * is now done in filters.c:block_url()
840 * - Made error conditions like domain lookup
841 * failiure or (various) problems while talking
842 * to the server use cgi.c:error_response()
843 * instead of generating HTML/HTTP in chat() (yuck!)
844 * - Removed logentry from cancelled commit
846 * Revision 1.20 2001/06/09 10:55:28 jongfoster
847 * Changing BUFSIZ ==> BUFFER_SIZE
849 * Revision 1.19 2001/06/07 23:12:52 jongfoster
850 * Replacing function pointer in struct gateway with a directly
851 * called function forwarded_connect().
852 * Replacing struct gateway with struct forward_spec
854 * Revision 1.18 2001/06/03 19:12:16 oes
855 * introduced new cgi handling
857 * Revision 1.17 2001/06/01 20:07:23 jongfoster
858 * Now uses action +image-blocker{} rather than config->tinygif
860 * Revision 1.16 2001/06/01 18:49:17 jongfoster
861 * Replaced "list_share" with "list" - the tiny memory gain was not
862 * worth the extra complexity.
864 * Revision 1.15 2001/05/31 21:24:47 jongfoster
865 * Changed "permission" to "action" throughout.
866 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
867 * Moved vanilla wafer check into chat(), since we must now
868 * decide whether or not to add it based on the URL.
870 * Revision 1.14 2001/05/29 20:14:01 joergs
871 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
874 * Revision 1.13 2001/05/29 09:50:24 jongfoster
875 * Unified blocklist/imagelist/permissionslist.
876 * File format is still under discussion, but the internal changes
879 * Also modified interceptor behaviour:
880 * - We now intercept all URLs beginning with one of the following
881 * prefixes (and *only* these prefixes):
883 * * http://ijbswa.sf.net/config/
884 * * http://ijbswa.sourceforge.net/config/
885 * - New interceptors "home page" - go to http://i.j.b/ to see it.
886 * - Internal changes so that intercepted and fast redirect pages
887 * are not replaced with an image.
888 * - Interceptors now have the option to send a binary page direct
889 * to the client. (i.e. ijb-send-banner uses this)
890 * - Implemented show-url-info interceptor. (Which is why I needed
891 * the above interceptors changes - a typical URL is
892 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
893 * The previous mechanism would not have intercepted that, and
894 * if it had been intercepted then it then it would have replaced
897 * Revision 1.12 2001/05/27 22:17:04 oes
899 * - re_process_buffer no longer writes the modified buffer
900 * to the client, which was very ugly. It now returns the
901 * buffer, which it is then written by chat.
903 * - content_length now adjusts the Content-Length: header
904 * for modified documents rather than crunch()ing it.
905 * (Length info in csp->content_length, which is 0 for
906 * unmodified documents)
908 * - For this to work, sed() is called twice when filtering.
910 * Revision 1.11 2001/05/26 17:27:53 jongfoster
911 * Added support for CLF and fixed LOG_LEVEL_LOG.
912 * Also did CRLF->LF fix of my previous patch.
914 * Revision 1.10 2001/05/26 15:26:15 jongfoster
915 * ACL feature now provides more security by immediately dropping
916 * connections from untrusted hosts.
918 * Revision 1.9 2001/05/26 00:28:36 jongfoster
919 * Automatic reloading of config file.
920 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
921 * Most of the global variables have been moved to a new
922 * struct configuration_spec, accessed through csp->config->globalname
923 * Most of the globals remaining are used by the Win32 GUI.
925 * Revision 1.8 2001/05/25 22:43:18 jongfoster
926 * Fixing minor memory leak and buffer overflow.
928 * Revision 1.7 2001/05/25 22:34:30 jongfoster
931 * Revision 1.6 2001/05/23 00:13:58 joergs
932 * AmigaOS support fixed.
934 * Revision 1.5 2001/05/22 18:46:04 oes
936 * - Enabled filtering banners by size rather than URL
937 * by adding patterns that replace all standard banner
938 * sizes with the "Junkbuster" gif to the re_filterfile
940 * - Enabled filtering WebBugs by providing a pattern
941 * which kills all 1x1 images
943 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
944 * which is selected by the (nonstandard and therefore
945 * capital) letter 'U' in the option string.
946 * It causes the quantifiers to be ungreedy by default.
947 * Appending a ? turns back to greedy (!).
949 * - Added a new interceptor ijb-send-banner, which
950 * sends back the "Junkbuster" gif. Without imagelist or
951 * MSIE detection support, or if tinygif = 1, or the
952 * URL isn't recognized as an imageurl, a lame HTML
953 * explanation is sent instead.
955 * - Added new feature, which permits blocking remote
956 * script redirects and firing back a local redirect
958 * The feature is conditionally compiled, i.e. it
959 * can be disabled with --disable-fast-redirects,
960 * plus it must be activated by a "fast-redirects"
961 * line in the config file, has its own log level
962 * and of course wants to be displayed by show-proxy-args
963 * Note: Boy, all the #ifdefs in 1001 locations and
964 * all the fumbling with configure.in and acconfig.h
965 * were *way* more work than the feature itself :-(
967 * - Because a generic redirect template was needed for
968 * this, tinygif = 3 now uses the same.
970 * - Moved GIFs, and other static HTTP response templates
975 * - Removed some >400 CRs again (Jon, you really worked
978 * Revision 1.4 2001/05/21 19:34:01 jongfoster
979 * Made failure to bind() a fatal error.
981 * Revision 1.3 2001/05/20 01:21:20 jongfoster
982 * Version 2.9.4 checkin.
983 * - Merged popupfile and cookiefile, and added control over PCRS
984 * filtering, in new "permissionsfile".
985 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
986 * file error you now get a message box (in the Win32 GUI) rather
987 * than the program exiting with no explanation.
988 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
990 * - Removed tabs from "config"
991 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
992 * - Bumped up version number.
994 * Revision 1.2 2001/05/17 22:34:44 oes
995 * - Added hint on GIF char array generation to jcc.c
996 * - Cleaned CRLF's from the sources and related files
997 * - Repaired logging for REF and FRC
999 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1000 * Initial import of version 2.9.3 source tree
1003 *********************************************************************/
1009 #include <sys/types.h>
1018 # ifndef FEATURE_PTHREAD
1022 # include <windows.h>
1023 # include <process.h>
1024 # endif /* ndef FEATURE_PTHREAD */
1027 # ifndef _WIN_CONSOLE
1028 # include "w32log.h"
1029 # endif /* ndef _WIN_CONSOLE */
1030 # include "w32svrapi.h"
1032 #else /* ifndef _WIN32 */
1034 # if !defined (__OS2__)
1035 # include <unistd.h>
1036 # include <sys/wait.h>
1037 # endif /* ndef __OS2__ */
1038 # include <sys/time.h>
1039 # include <sys/stat.h>
1040 # include <sys/ioctl.h>
1043 #include <sys/termios.h>
1051 # include <signal.h>
1054 # include <socket.h> /* BeOS has select() for sockets only. */
1055 # include <OS.h> /* declarations for threads and stuff. */
1058 # if defined(__EMX__) || defined(__OS2__)
1059 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1064 #define bzero(B,N) memset(B,0x00,n)
1068 # include <select.h>
1073 #include "project.h"
1076 #include "filters.h"
1077 #include "loaders.h"
1078 #include "parsers.h"
1079 #include "miscutil.h"
1081 #include "jbsockets.h"
1082 #include "gateway.h"
1083 #include "actions.h"
1085 #include "loadcfg.h"
1086 #include "urlmatch.h"
1088 const char jcc_h_rcs[] = JCC_H_VERSION;
1089 const char project_h_rcs[] = PROJECT_H_VERSION;
1092 struct client_state clients[1];
1093 struct file_list files[1];
1095 #ifdef FEATURE_STATISTICS
1096 int urls_read = 0; /* total nr of urls read inc rejected */
1097 int urls_rejected = 0; /* total nr of urls rejected */
1098 #endif /* def FEATURE_STATISTICS */
1100 #ifdef FEATURE_GRACEFUL_TERMINATION
1101 int g_terminate = 0;
1104 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1105 static void sig_handler(int the_signal);
1107 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1108 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1109 static jb_err get_server_headers(struct client_state *csp);
1110 static const char *crunch_reason(const struct http_response *rsp);
1111 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
1113 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1115 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1116 static jb_err change_request_destination(struct client_state *csp);
1117 static void chat(struct client_state *csp);
1118 static void serve(struct client_state *csp);
1119 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1120 static void usage(const char *myname);
1122 static void initialize_mutexes(void);
1123 static jb_socket bind_port_helper(struct configuration_spec *config);
1124 static void listen_loop(void);
1127 void serve(struct client_state *csp);
1128 #else /* ifndef AMIGA */
1129 static void serve(struct client_state *csp);
1130 #endif /* def AMIGA */
1133 static int32 server_thread(void *data);
1134 #endif /* def __BEOS__ */
1137 #define sleep(N) Sleep(((N) * 1000))
1141 #define sleep(N) DosSleep(((N) * 100))
1144 #ifdef FEATURE_PTHREAD
1145 pthread_mutex_t log_mutex;
1146 pthread_mutex_t log_init_mutex;
1148 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1149 pthread_mutex_t resolver_mutex;
1150 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1152 #ifndef HAVE_GMTIME_R
1153 pthread_mutex_t gmtime_mutex;
1154 #endif /* ndef HAVE_GMTIME_R */
1156 #ifndef HAVE_LOCALTIME_R
1157 pthread_mutex_t localtime_mutex;
1158 #endif /* ndef HAVE_GMTIME_R */
1161 pthread_mutex_t rand_mutex;
1162 #endif /* ndef HAVE_RANDOM */
1164 #endif /* FEATURE_PTHREAD */
1166 #if defined(unix) || defined(__EMX__)
1167 const char *basedir = NULL;
1168 const char *pidfile = NULL;
1169 int received_hup_signal = 0;
1170 #endif /* defined unix */
1172 /* HTTP snipplets. */
1173 static const char CSUCCEED[] =
1174 "HTTP/1.0 200 Connection established\r\n"
1175 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1177 static const char CHEADER[] =
1178 "HTTP/1.0 400 Invalid header received from client\r\n"
1179 "Proxy-Agent: Privoxy " VERSION "\r\n"
1180 "Content-Type: text/plain\r\n"
1181 "Connection: close\r\n\r\n"
1182 "Invalid header received from client.\r\n";
1184 static const char FTP_RESPONSE[] =
1185 "HTTP/1.0 400 Invalid request received from client\r\n"
1186 "Content-Type: text/plain\r\n"
1187 "Connection: close\r\n\r\n"
1188 "Invalid request. Privoxy doesn't support FTP.\r\n";
1190 static const char GOPHER_RESPONSE[] =
1191 "HTTP/1.0 400 Invalid request received from client\r\n"
1192 "Content-Type: text/plain\r\n"
1193 "Connection: close\r\n\r\n"
1194 "Invalid request. Privoxy doesn't support gopher.\r\n";
1196 /* XXX: should be a template */
1197 static const char MISSING_DESTINATION_RESPONSE[] =
1198 "HTTP/1.0 400 Bad request received from client\r\n"
1199 "Proxy-Agent: Privoxy " VERSION "\r\n"
1200 "Content-Type: text/plain\r\n"
1201 "Connection: close\r\n\r\n"
1202 "Bad request. Privoxy was unable to extract the destination.\r\n";
1204 /* XXX: should be a template */
1205 static const char NO_SERVER_DATA_RESPONSE[] =
1206 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1207 "Proxy-Agent: Privoxy " VERSION "\r\n"
1208 "Content-Type: text/plain\r\n"
1209 "Connection: close\r\n\r\n"
1210 "Empty server or forwarder response.\r\n"
1211 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1213 /* XXX: should be a template */
1214 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1215 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1216 "Proxy-Agent: Privoxy " VERSION "\r\n"
1217 "Content-Type: text/plain\r\n"
1218 "Connection: close\r\n\r\n"
1219 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1222 /* XXX: should be a template */
1223 static const char NULL_BYTE_RESPONSE[] =
1224 "HTTP/1.0 400 Bad request received from client\r\n"
1225 "Proxy-Agent: Privoxy " VERSION "\r\n"
1226 "Content-Type: text/plain\r\n"
1227 "Connection: close\r\n\r\n"
1228 "Bad request. Null byte(s) before end of request.\r\n";
1231 /* XXX: should be a template */
1232 static const char MESSED_UP_REQUEST_RESPONSE[] =
1233 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1234 "Proxy-Agent: Privoxy " VERSION "\r\n"
1235 "Content-Type: text/plain\r\n"
1236 "Connection: close\r\n\r\n"
1237 "Bad request. Messed up with header filters.\r\n";
1239 /* A function to crunch a response */
1240 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1242 /* Crunch function flags */
1243 #define CF_NO_FLAGS 0
1244 /* Cruncher applies to forced requests as well */
1245 #define CF_IGNORE_FORCE 1
1246 /* Crunched requests are counted for the block statistics */
1247 #define CF_COUNT_AS_REJECT 2
1249 /* A crunch function and its flags */
1252 const crunch_func_ptr cruncher;
1256 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1258 /* Complete list of cruncher functions */
1259 static const struct cruncher crunchers_all[] = {
1260 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1261 { block_url, CF_COUNT_AS_REJECT },
1262 #ifdef FEATURE_TRUST
1263 { trust_url, CF_COUNT_AS_REJECT },
1264 #endif /* def FEATURE_TRUST */
1265 { redirect_url, CF_NO_FLAGS },
1266 { dispatch_cgi, CF_IGNORE_FORCE},
1270 /* Light version, used after tags are applied */
1271 static const struct cruncher crunchers_light[] = {
1272 { block_url, CF_COUNT_AS_REJECT },
1273 { redirect_url, CF_NO_FLAGS },
1278 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1279 /*********************************************************************
1281 * Function : sig_handler
1283 * Description : Signal handler for different signals.
1284 * Exit gracefully on TERM and INT
1285 * or set a flag that will cause the errlog
1286 * to be reopened by the main thread on HUP.
1289 * 1 : the_signal = the signal cause this function to call
1293 *********************************************************************/
1294 static void sig_handler(int the_signal)
1300 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1311 received_hup_signal = 1;
1316 * We shouldn't be here, unless we catch signals
1317 * in main() that we can't handle here!
1319 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1327 /*********************************************************************
1329 * Function : client_protocol_is_unsupported
1331 * Description : Checks if the client used a known unsupported
1332 * protocol and deals with it by sending an error
1336 * 1 : csp = Current client state (buffers, headers, etc...)
1337 * 2 : req = the first request line send by the client
1339 * Returns : TRUE if an error response has been generated, or
1340 * FALSE if the request doesn't look invalid.
1342 *********************************************************************/
1343 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1346 * If it's a FTP or gopher request, we don't support it.
1348 * These checks are better than nothing, but they might
1349 * not work in all configurations and some clients might
1350 * have problems digesting the answer.
1352 * They should, however, never cause more problems than
1353 * Privoxy's old behaviour (returning the misleading HTML
1356 * "Could not resolve http://(ftp|gopher)://example.org").
1358 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1360 const char *response = NULL;
1361 const char *protocol = NULL;
1363 if (!strncmpic(req, "GET ftp://", 10))
1365 response = FTP_RESPONSE;
1370 response = GOPHER_RESPONSE;
1371 protocol = "GOPHER";
1373 log_error(LOG_LEVEL_ERROR,
1374 "%s tried to use Privoxy as %s proxy: %s",
1375 csp->ip_addr_str, protocol, req);
1376 log_error(LOG_LEVEL_CLF,
1377 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1379 write_socket(csp->cfd, response, strlen(response));
1388 /*********************************************************************
1390 * Function : get_request_destination_elsewhere
1392 * Description : If the client's request was redirected into
1393 * Privoxy without the client's knowledge,
1394 * the request line lacks the destination host.
1396 * This function tries to get it elsewhere,
1397 * provided accept-intercepted-requests is enabled.
1399 * "Elsewhere" currently only means "Host: header",
1400 * but in the future we may ask the redirecting
1401 * packet filter to look the destination up.
1403 * If the destination stays unknown, an error
1404 * response is send to the client and headers
1405 * are freed so that chat() can return directly.
1408 * 1 : csp = Current client state (buffers, headers, etc...)
1409 * 2 : headers = a header list
1411 * Returns : JB_ERR_OK if the destination is now known, or
1412 * JB_ERR_PARSE if it isn't.
1414 *********************************************************************/
1415 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1419 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1421 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1422 " Privoxy isn't configured to accept intercepted requests.",
1423 csp->ip_addr_str, csp->http->cmd);
1424 /* XXX: Use correct size */
1425 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1426 csp->ip_addr_str, csp->http->cmd);
1428 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1429 destroy_list(headers);
1431 return JB_ERR_PARSE;
1433 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1435 /* Split the domain we just got for pattern matching */
1436 init_domain_components(csp->http);
1442 /* We can't work without destination. Go spread the news.*/
1444 req = list_to_text(headers);
1446 /* XXX: Use correct size */
1447 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1448 csp->ip_addr_str, csp->http->cmd);
1449 log_error(LOG_LEVEL_ERROR,
1450 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1451 csp->ip_addr_str, csp->http->cmd, req);
1454 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1455 destroy_list(headers);
1457 return JB_ERR_PARSE;
1460 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1461 * to get the destination IP address, use it as host directly
1462 * or do a reverse DNS lookup first.
1467 /*********************************************************************
1469 * Function : get_server_headers
1471 * Description : Parses server headers in iob and fills them
1472 * into csp->headers so that they can later be
1476 * 1 : csp = Current client state (buffers, headers, etc...)
1478 * Returns : JB_ERR_OK if everything went fine, or
1479 * JB_ERR_PARSE if the headers were incomplete.
1481 *********************************************************************/
1482 static jb_err get_server_headers(struct client_state *csp)
1484 int continue_hack_in_da_house = 0;
1487 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1492 * continue hack in da house. Ignore the ending of
1493 * this head and continue enlisting header lines.
1494 * The reason is described below.
1496 enlist(csp->headers, "");
1497 continue_hack_in_da_house = 0;
1500 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1503 * It's a bodyless continue response, don't
1504 * stop header parsing after reaching it's end.
1506 * As a result Privoxy will concatenate the
1507 * next response's head and parse and deliver
1508 * the headers as if they belonged to one request.
1510 * The client will separate them because of the
1511 * empty line between them.
1513 * XXX: What we're doing here is clearly against
1514 * the intended purpose of the continue header,
1515 * and under some conditions (HTTP/1.0 client request)
1516 * it's a standard violation.
1518 * Anyway, "sort of against the spec" is preferable
1519 * to "always getting confused by Continue responses"
1520 * (Privoxy's behaviour before this hack was added)
1522 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1523 continue_hack_in_da_house = 1;
1525 else if (*header == '\0')
1528 * If the header is empty, but the Continue hack
1529 * isn't active, we can assume that we reached the
1530 * end of the buffer before we hit the end of the
1533 * Inform the caller an let it decide how to handle it.
1535 return JB_ERR_PARSE;
1538 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1541 * XXX: Should we quit the request and return a
1542 * out of memory error page instead?
1544 log_error(LOG_LEVEL_ERROR,
1545 "Out of memory while enlisting server headers. %s lost.",
1555 /*********************************************************************
1557 * Function : crunch_reason
1559 * Description : Translates the crunch reason code into a string.
1562 * 1 : rsp = a http_response
1564 * Returns : A string with the crunch reason or an error description.
1566 *********************************************************************/
1567 static const char *crunch_reason(const struct http_response *rsp)
1569 char * reason = NULL;
1571 assert(rsp != NULL);
1574 return "Internal error while searching for crunch reason";
1577 switch (rsp->reason)
1579 case RSP_REASON_UNSUPPORTED:
1580 reason = "Unsupported HTTP feature";
1582 case RSP_REASON_BLOCKED:
1585 case RSP_REASON_UNTRUSTED:
1586 reason = "Untrusted";
1588 case RSP_REASON_REDIRECTED:
1589 reason = "Redirected";
1591 case RSP_REASON_CGI_CALL:
1592 reason = "CGI Call";
1594 case RSP_REASON_NO_SUCH_DOMAIN:
1595 reason = "DNS failure";
1597 case RSP_REASON_FORWARDING_FAILED:
1598 reason = "Forwarding failed";
1600 case RSP_REASON_CONNECT_FAILED:
1601 reason = "Connection failure";
1603 case RSP_REASON_OUT_OF_MEMORY:
1604 reason = "Out of memory (may mask other reasons)";
1607 reason = "No reason recorded";
1615 /*********************************************************************
1617 * Function : send_crunch_response
1619 * Description : Delivers already prepared response for
1620 * intercepted requests, logs the interception
1621 * and frees the response.
1624 * 1 : csp = Current client state (buffers, headers, etc...)
1625 * 1 : rsp = Fully prepared response. Will be freed on exit.
1627 * Returns : Nothing.
1629 *********************************************************************/
1630 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1632 const struct http_request *http = csp->http;
1633 char status_code[4];
1635 assert(rsp != NULL);
1636 assert(rsp->head != NULL);
1641 * Not supposed to happen. If it does
1642 * anyway, treat it as an unknown error.
1644 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1645 /* return code doesn't matter */
1650 /* If rsp is still NULL, we have serious internal problems. */
1651 log_error(LOG_LEVEL_FATAL,
1652 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1656 * Extract the status code from the actual head
1657 * that was send to the client. It is the only
1658 * way to get it right for all requests, including
1659 * the fixed ones for out-of-memory problems.
1661 * A head starts like this: 'HTTP/1.1 200...'
1665 status_code[0] = rsp->head[9];
1666 status_code[1] = rsp->head[10];
1667 status_code[2] = rsp->head[11];
1668 status_code[3] = '\0';
1670 /* Write the answer to the client */
1671 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1672 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1674 /* There is nothing we can do about it. */
1675 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1678 /* Log that the request was crunched and why. */
1679 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1680 http->hostport, http->path, crunch_reason(rsp));
1681 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1682 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1684 /* Clean up and return */
1685 if (cgi_error_memory() != rsp)
1687 free_http_response(rsp);
1694 /*********************************************************************
1696 * Function : request_contains_null_bytes
1698 * Description : Checks for NULL bytes in the request and sends
1699 * an error message to the client if any were found.
1701 * XXX: currently not used, see comment in chat().
1704 * 1 : csp = Current client state (buffers, headers, etc...)
1705 * 2 : buf = Data from the client's request to check.
1706 * 3 : len = The data length.
1708 * Returns : TRUE if the request contained one or more NULL bytes, or
1711 *********************************************************************/
1712 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1714 size_t c_len; /* Request lenght when treated as C string */
1716 c_len = strlen(buf);
1721 * Null byte(s) found. Log the request,
1722 * return an error response and hang up.
1724 size_t tmp_len = c_len;
1729 * Replace NULL byte(s) with '°' characters
1730 * so the request can be logged as string.
1731 * XXX: Is there a better replacement character?
1734 tmp_len += strlen(buf+tmp_len);
1735 } while (tmp_len < len);
1737 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1738 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1739 log_error(LOG_LEVEL_HEADER,
1740 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1742 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1744 /* XXX: Log correct size */
1745 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1755 /*********************************************************************
1757 * Function : crunch_response_triggered
1759 * Description : Checks if the request has to be crunched,
1760 * and delivers the crunch response if necessary.
1763 * 1 : csp = Current client state (buffers, headers, etc...)
1764 * 2 : crunchers = list of cruncher functions to run
1766 * Returns : TRUE if the request was answered with a crunch response
1769 *********************************************************************/
1770 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1772 struct http_response *rsp = NULL;
1773 const struct cruncher *c;
1776 * If CGI request crunching is disabled,
1777 * check the CGI dispatcher out of order to
1778 * prevent unintentional blocks or redirects.
1780 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1781 && (NULL != (rsp = dispatch_cgi(csp))))
1783 /* Deliver, log and free the interception response. */
1784 send_crunch_response(csp, rsp);
1788 for (c = crunchers; c->cruncher != NULL; c++)
1791 * Check the cruncher if either Privoxy is toggled
1792 * on and the request isn't forced, or if the cruncher
1793 * applies to forced requests as well.
1795 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1796 !(csp->flags & CSP_FLAG_FORCED)) ||
1797 (c->flags & CF_IGNORE_FORCE))
1799 rsp = c->cruncher(csp);
1802 /* Deliver, log and free the interception response. */
1803 send_crunch_response(csp, rsp);
1804 #ifdef FEATURE_STATISTICS
1805 if (c->flags & CF_COUNT_AS_REJECT)
1807 csp->flags |= CSP_FLAG_REJECTED;
1809 #endif /* def FEATURE_STATISTICS */
1820 /*********************************************************************
1822 * Function : build_request_line
1824 * Description : Builds the HTTP request line.
1826 * If a HTTP forwarder is used it expects the whole URL,
1827 * web servers only get the path.
1830 * 1 : csp = Current client state (buffers, headers, etc...)
1831 * 2 : fwd = The forwarding spec used for the request
1832 * XXX: Should use http->fwd instead.
1833 * 3 : request_line = The old request line which will be replaced.
1835 * Returns : Nothing. Terminates in case of memory problems.
1837 *********************************************************************/
1838 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1840 struct http_request *http = csp->http;
1842 assert(http->ssl == 0);
1845 * Downgrade http version from 1.1 to 1.0
1846 * if +downgrade action applies.
1848 if ( (csp->action->flags & ACTION_DOWNGRADE)
1849 && (!strcmpic(http->ver, "HTTP/1.1")))
1852 http->ver = strdup("HTTP/1.0");
1854 if (http->ver == NULL)
1856 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1861 * Rebuild the request line.
1863 freez(*request_line);
1864 *request_line = strdup(http->gpc);
1865 string_append(request_line, " ");
1867 if (fwd->forward_host)
1869 string_append(request_line, http->url);
1873 string_append(request_line, http->path);
1875 string_append(request_line, " ");
1876 string_append(request_line, http->ver);
1878 if (*request_line == NULL)
1880 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1882 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1886 /*********************************************************************
1888 * Function : change_request_destination
1890 * Description : Parse a (rewritten) request line and regenerate
1891 * the http request data.
1894 * 1 : csp = Current client state (buffers, headers, etc...)
1896 * Returns : Forwards the parse_http_request() return code.
1897 * Terminates in case of memory problems.
1899 *********************************************************************/
1900 static jb_err change_request_destination(struct client_state *csp)
1902 struct http_request *http = csp->http;
1905 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1906 free_http_request(http);
1907 err = parse_http_request(csp->headers->first->str, http, csp);
1908 if (JB_ERR_OK != err)
1910 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1911 jb_err_to_string(err));
1913 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1914 if (http->ocmd == NULL)
1916 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1923 /*********************************************************************
1927 * Description : Once a connection to the client has been accepted,
1928 * this function is called (via serve()) to handle the
1929 * main business of the communication. When this
1930 * function returns, the caller must close the client
1933 * FIXME: chat is nearly thousand lines long.
1937 * 1 : csp = Current client state (buffers, headers, etc...)
1939 * Returns : Nothing.
1941 *********************************************************************/
1942 static void chat(struct client_state *csp)
1944 char buf[BUFFER_SIZE];
1952 int ms_iis5_hack = 0;
1953 size_t byte_count = 0;
1954 int forwarded_connect_retries = 0;
1955 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1956 const struct forward_spec * fwd;
1957 struct http_request *http;
1958 int len; /* for buffer sizes (and negative error codes) */
1961 /* Function that does the content filtering for the current request */
1962 filter_function_ptr content_filter = NULL;
1964 /* Skeleton for HTTP response, if we should intercept the request */
1965 struct http_response *rsp;
1967 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1968 struct list header_list;
1969 struct list *headers = &header_list;
1973 memset(buf, 0, sizeof(buf));
1976 * Read the client's request. Note that since we're not using select() we
1977 * could get blocked here if a client connected, then didn't say anything!
1982 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1984 if (len <= 0) break; /* error! */
1987 * If there is no memory left for buffering the
1988 * request, there is nothing we can do but hang up
1990 if (add_to_iob(csp, buf, len))
1995 req = get_header(csp->iob);
1997 } while ((NULL != req) && ('\0' == *req));
1999 if ((NULL != req) && ('\0' != *req))
2001 /* Request received. Validate and parse it. */
2005 * XXX: Temporary disabled to prevent problems
2006 * with POST requests whose bodies are allowed to
2007 * contain NULL bytes. BR#1730105.
2009 * The main purpose of this check is to properly
2010 * log stuff like BitTorrent traffic and other junk
2011 * that hits public proxies. It's not required for
2012 * Privoxy to functions as those requests are discarded
2015 * It probably should be rewritten to only check
2016 * the head of the request. Another option would
2017 * be to let all POST requests pass, although that
2018 * may not be good enough.
2020 if (request_contains_null_bytes(csp, buf, len))
2022 /* NULL bytes found and dealt with, just hang up. */
2027 /* Does the request line look invalid? */
2028 if (client_protocol_is_unsupported(csp, req))
2031 * Yes. The request has already been
2032 * answered with a error response, the buffers
2033 * were freed and we're done with chatting.
2038 #ifdef FEATURE_FORCE_LOAD
2040 * If this request contains the FORCE_PREFIX and blocks
2041 * aren't enforced, get rid of it and set the force flag.
2043 if (strstr(req, FORCE_PREFIX))
2045 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2047 log_error(LOG_LEVEL_FORCE,
2048 "Ignored force prefix in request: \"%s\".", req);
2052 strclean(req, FORCE_PREFIX);
2053 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2054 csp->flags |= CSP_FLAG_FORCED;
2058 #endif /* def FEATURE_FORCE_LOAD */
2059 err = parse_http_request(req, http, csp);
2060 if (JB_ERR_OK != err)
2062 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2068 if (http->cmd == NULL)
2070 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2071 /* XXX: Use correct size */
2072 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2073 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2075 free_http_request(http);
2079 /* grab the rest of the client's headers */
2083 p = get_header(csp->iob);
2087 /* There are no additional headers to read. */
2094 * We didn't receive a complete header
2095 * line yet, get the rest of it.
2097 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2100 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2101 destroy_list(headers);
2105 if (add_to_iob(csp, buf, len))
2108 * If there is no memory left for buffering the
2109 * request, there is nothing we can do but hang up
2111 destroy_list(headers);
2118 * We were able to read a complete
2119 * header and can finaly enlist it.
2126 if (http->host == NULL)
2129 * If we still don't know the request destination,
2130 * the request is invalid or the client uses
2131 * Privoxy without its knowledge.
2133 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2136 * Our attempts to get the request destination
2137 * elsewhere failed or Privoxy is configured
2138 * to only accept proxy requests.
2140 * An error response has already been send
2141 * and we're done here.
2148 * Determine the actions for this URL
2150 #ifdef FEATURE_TOGGLE
2151 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2153 /* Most compatible set of actions (i.e. none) */
2154 init_current_action(csp->action);
2157 #endif /* ndef FEATURE_TOGGLE */
2159 get_url_actions(csp, http);
2163 * Save a copy of the original request for logging
2165 http->ocmd = strdup(http->cmd);
2167 if (http->ocmd == NULL)
2169 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2172 enlist(csp->headers, http->cmd);
2174 /* Append the previously read headers */
2175 list_append_list_unique(csp->headers, headers);
2176 destroy_list(headers);
2178 err = sed(client_patterns, add_client_headers, csp);
2179 if (JB_ERR_OK != err)
2181 assert(err == JB_ERR_PARSE);
2182 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2184 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2186 /* Check request line for rewrites. */
2187 if ((NULL == csp->headers->first->str)
2188 || (strcmp(http->cmd, csp->headers->first->str) &&
2189 (JB_ERR_OK != change_request_destination(csp))))
2192 * A header filter broke the request line - bail out.
2194 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2195 /* XXX: Use correct size */
2196 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2197 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2199 free_http_request(http);
2203 /* decide how to route the HTTP request */
2204 fwd = forward_url(csp, http);
2207 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2208 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2212 * build the http request to send to the server
2213 * we have to do one of the following:
2215 * create = use the original HTTP request to create a new
2216 * HTTP request that has either the path component
2217 * without the http://domainspec (w/path) or the
2218 * full orininal URL (w/url)
2219 * Note that the path and/or the HTTP version may
2220 * have been altered by now.
2222 * connect = Open a socket to the host:port of the server
2223 * and short-circuit server and client socket.
2225 * pass = Pass the request unchanged if forwarding a CONNECT
2226 * request to a parent proxy. Note that we'll be sending
2227 * the CFAIL message ourselves if connecting to the parent
2228 * fails, but we won't send a CSUCCEED message if it works,
2229 * since that would result in a double message (ours and the
2230 * parent's). After sending the request to the parent, we simply
2233 * here's the matrix:
2236 * +--------+--------+
2238 * 0 | create | connect|
2240 * Forwarding +--------+--------+
2242 * 1 | create | pass |
2244 * +--------+--------+
2248 if (http->ssl && connect_port_is_forbidden(csp))
2250 const char *acceptable_connect_ports =
2251 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2252 assert(NULL != acceptable_connect_ports);
2253 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2254 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2255 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2256 csp->action->flags |= ACTION_BLOCK;
2262 freez(csp->headers->first->str);
2263 build_request_line(csp, fwd, &csp->headers->first->str);
2266 hdr = list_to_text(csp->headers);
2269 /* FIXME Should handle error properly */
2270 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2274 * We have a request. Check if one of the crunchers wants it.
2276 if (crunch_response_triggered(csp, crunchers_all))
2279 * Yes. The client got the crunch response
2280 * and we are done here after cleaning up.
2283 list_remove_all(csp->headers);
2289 * The headers can't be removed earlier because
2290 * they were still needed for the referrer check
2291 * in case of CGI crunches.
2293 * XXX: Would it be worth to move the referrer check
2294 * into client_referrer() and set a flag if it's trusted?
2296 list_remove_all(csp->headers);
2298 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2300 if (fwd->forward_host)
2302 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2303 fwd->forward_host, fwd->forward_port, http->hostport);
2307 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2310 /* here we connect to the server, gateway, or the forwarder */
2312 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2313 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2315 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2316 forwarded_connect_retries, http->hostport);
2319 if (csp->sfd == JB_INVALID_SOCKET)
2321 if (fwd->type != SOCKS_NONE)
2324 rsp = error_response(csp, "forwarding-failed", errno);
2326 else if (errno == EINVAL)
2328 rsp = error_response(csp, "no-such-domain", errno);
2332 rsp = error_response(csp, "connect-failed", errno);
2333 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2338 /* Write the answer to the client */
2341 send_crunch_response(csp, rsp);
2348 if (fwd->forward_host || (http->ssl == 0))
2350 /* write the client's (modified) header to the server
2351 * (along with anything else that may be in the buffer)
2354 if (write_socket(csp->sfd, hdr, strlen(hdr))
2355 || (flush_socket(csp->sfd, csp->iob) < 0))
2357 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2360 rsp = error_response(csp, "connect-failed", errno);
2364 send_crunch_response(csp, rsp);
2374 * We're running an SSL tunnel and we're not forwarding,
2375 * so just send the "connect succeeded" message to the
2376 * client, flush the rest, and get out of the way.
2378 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2386 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2388 /* we're finished with the client's header */
2391 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2393 /* pass data between the client and server
2394 * until one or the other shuts down the connection.
2403 * FD_ZERO here seems to point to an errant macro which crashes.
2404 * So do this by hand for now...
2406 memset(&rfds,0x00,sizeof(fd_set));
2410 FD_SET(csp->cfd, &rfds);
2411 FD_SET(csp->sfd, &rfds);
2413 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2417 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2421 /* this is the body of the browser's request
2422 * just read it and write it.
2425 if (FD_ISSET(csp->cfd, &rfds))
2427 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2431 break; /* "game over, man" */
2434 if (write_socket(csp->sfd, buf, (size_t)len))
2436 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2443 * The server wants to talk. It could be the header or the body.
2444 * If `hdr' is null, then it's the header otherwise it's the body.
2445 * FIXME: Does `hdr' really mean `host'? No.
2449 if (FD_ISSET(csp->sfd, &rfds))
2452 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2456 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2458 if (http->ssl && (fwd->forward_host == NULL))
2461 * Just hang up. We already confirmed the client's CONNECT
2462 * request with status code 200 and unencrypted content is
2463 * no longer welcome.
2465 log_error(LOG_LEVEL_ERROR,
2466 "CONNECT already confirmed. Unable to tell the client about the problem.");
2469 else if (byte_count)
2472 * Just hang up. We already transmitted the original headers
2473 * and parts of the original content and therefore missed the
2474 * chance to send an error message (without risking data corruption).
2476 * XXX: we could retry with a fancy range request here.
2478 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2479 "Unable to tell the client about the problem.");
2483 rsp = error_response(csp, "connect-failed", errno);
2487 send_crunch_response(csp, rsp);
2493 /* Add a trailing zero. This lets filter_popups
2494 * use string operations.
2498 /* Normally, this would indicate that we've read
2499 * as much as the server has sent us and we can
2500 * close the client connection. However, Microsoft
2501 * in its wisdom has released IIS/5 with a bug that
2502 * prevents it from sending the trailing \r\n in
2503 * a 302 redirect header (and possibly other headers).
2504 * To work around this if we've haven't parsed
2505 * a full header we'll append a trailing \r\n
2506 * and see if this now generates a valid one.
2508 * This hack shouldn't have any impacts. If we've
2509 * already transmitted the header or if this is a
2510 * SSL connection, then we won't bother with this
2511 * hack. So we only work on partially received
2512 * headers. If we append a \r\n and this still
2513 * doesn't generate a valid header, then we won't
2514 * transmit anything to the client.
2519 if (server_body || http->ssl)
2522 * If we have been buffering up the document,
2523 * now is the time to apply content modification
2524 * and send the result to the client.
2528 p = execute_content_filter(csp, content_filter);
2530 * If the content filter fails, use the original
2531 * buffer and length.
2532 * (see p != NULL ? p : csp->iob->cur below)
2536 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2539 if (JB_ERR_OK != update_server_headers(csp))
2541 log_error(LOG_LEVEL_FATAL,
2542 "Failed to update server headers. after filtering.");
2545 hdr = list_to_text(csp->headers);
2548 /* FIXME Should handle error properly */
2549 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2552 if (write_socket(csp->cfd, hdr, strlen(hdr))
2553 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2555 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2565 break; /* "game over, man" */
2569 * This is NOT the body, so
2570 * Let's pretend the server just sent us a blank line.
2572 snprintf(buf, sizeof(buf), "\r\n");
2573 len = (int)strlen(buf);
2576 * Now, let the normal header parsing algorithm below do its
2577 * job. If it fails, we'll exit instead of continuing.
2584 * If this is an SSL connection or we're in the body
2585 * of the server document, just write it to the client,
2586 * unless we need to buffer the body for later content-filtering
2589 if (server_body || http->ssl)
2594 * If there is no memory left for buffering the content, or the buffer limit
2595 * has been reached, switch to non-filtering mode, i.e. make & write the
2596 * header, flush the iob and buf, and get out of the way.
2598 if (add_to_iob(csp, buf, len))
2603 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2605 hdr = list_to_text(csp->headers);
2609 * Memory is too tight to even generate the header.
2610 * Send our static "Out-of-memory" page.
2612 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2613 rsp = cgi_error_memory();
2614 send_crunch_response(csp, rsp);
2618 hdrlen = strlen(hdr);
2620 if (write_socket(csp->cfd, hdr, hdrlen)
2621 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2622 || (write_socket(csp->cfd, buf, (size_t)len)))
2624 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2631 * Reset the byte_count to the amount of bytes
2632 * we just flushed. len will be added a few lines below,
2633 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2635 byte_count = (size_t)flushed;
2637 content_filter = NULL;
2643 if (write_socket(csp->cfd, buf, (size_t)len))
2645 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2649 byte_count += (size_t)len;
2654 /* we're still looking for the end of the
2655 * server's header ... (does that make header
2656 * parsing an "out of body experience" ?
2660 * buffer up the data we just read. If that fails,
2661 * there's little we can do but send our static
2662 * out-of-memory page.
2664 if (add_to_iob(csp, buf, len))
2666 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2667 rsp = cgi_error_memory();
2668 send_crunch_response(csp, rsp);
2673 /* Convert iob into something sed() can digest */
2674 if (JB_ERR_PARSE == get_server_headers(csp))
2678 /* Well, we tried our MS IIS/5
2679 * hack and it didn't work.
2680 * The header is incomplete
2681 * and there isn't anything
2682 * we can do about it.
2684 log_error(LOG_LEVEL_INFO,
2685 "MS IIS5 hack didn't produce valid headers.");
2690 /* Since we have to wait for
2691 * more from the server before
2692 * we can parse the headers
2693 * we just continue here.
2699 /* Did we actually get anything? */
2700 if (NULL == csp->headers->first)
2702 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2703 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2704 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2705 free_http_request(http);
2709 assert(csp->headers->first->str);
2711 if (strncmpic(csp->headers->first->str, "HTTP", 4))
2714 * It doesn't look like a HTTP response:
2715 * tell the client and log the problem.
2717 if (strlen(csp->headers->first->str) > 30)
2719 csp->headers->first->str[30] = '\0';
2721 log_error(LOG_LEVEL_ERROR,
2722 "Invalid server or forwarder response. Starts with: %s",
2723 csp->headers->first->str);
2724 log_error(LOG_LEVEL_CLF,
2725 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2726 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2727 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2728 free_http_request(http);
2732 /* we have now received the entire header.
2733 * filter it and send the result to the client
2735 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2737 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2739 hdr = list_to_text(csp->headers);
2742 /* FIXME Should handle error properly */
2743 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2746 if (crunch_response_triggered(csp, crunchers_light))
2749 * One of the tags created by a server-header
2750 * tagger triggered a crunch. We already
2751 * delivered the crunch response to the client
2752 * and are done here after cleaning up.
2757 /* Buffer and pcrs filter this if appropriate. */
2759 if (!http->ssl) /* We talk plaintext */
2761 content_filter = get_filter_function(csp);
2764 * Only write if we're not buffering for content modification
2766 if (!content_filter)
2768 /* write the server's (modified) header to
2769 * the client (along with anything else that
2770 * may be in the buffer)
2773 if (write_socket(csp->cfd, hdr, strlen(hdr))
2774 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2776 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2778 /* the write failed, so don't bother
2779 * mentioning it to the client...
2780 * it probably can't hear us anyway.
2786 byte_count += (size_t)len;
2789 /* we're finished with the server's header */
2794 /* If this was a MS IIS/5 hack then it means
2795 * the server has already closed the
2796 * connection. Nothing more to read. Time
2801 log_error(LOG_LEVEL_INFO,
2802 "Closed server connection detected with MS IIS5 hack enabled.");
2809 return; /* huh? we should never get here */
2812 if (csp->content_length == 0)
2815 * If Privoxy didn't recalculate the
2816 * Content-Lenght, byte_count is still
2819 csp->content_length = byte_count;
2822 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2823 csp->ip_addr_str, http->ocmd, csp->content_length);
2827 /*********************************************************************
2831 * Description : This is little more than chat. We only "serve" to
2832 * to close any socket that chat may have opened.
2835 * 1 : csp = Current client state (buffers, headers, etc...)
2839 *********************************************************************/
2841 void serve(struct client_state *csp)
2842 #else /* ifndef AMIGA */
2843 static void serve(struct client_state *csp)
2844 #endif /* def AMIGA */
2847 close_socket(csp->cfd);
2849 if (csp->sfd != JB_INVALID_SOCKET)
2851 close_socket(csp->sfd);
2854 csp->flags &= ~CSP_FLAG_ACTIVE;
2860 /*********************************************************************
2862 * Function : server_thread
2864 * Description : We only exist to call `serve' in a threaded environment.
2867 * 1 : data = Current client state (buffers, headers, etc...)
2869 * Returns : Always 0.
2871 *********************************************************************/
2872 static int32 server_thread(void *data)
2874 serve((struct client_state *) data);
2881 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2882 /*********************************************************************
2886 * Description : Print usage info & exit.
2888 * Parameters : Pointer to argv[0] for identifying ourselves
2892 *********************************************************************/
2893 static void usage(const char *myname)
2895 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2899 #endif /* defined(unix) */
2902 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2903 #endif /* defined(unix) */
2904 "[--version] [configfile]\n"
2905 "Aborting\n", myname);
2910 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2913 /*********************************************************************
2915 * Function : initialize_mutexes
2917 * Description : Prepares mutexes if mutex support is available.
2921 * Returns : Void, exits in case of errors.
2923 *********************************************************************/
2924 static void initialize_mutexes(void)
2928 #ifdef FEATURE_PTHREAD
2930 * Prepare global mutex semaphores
2932 err = pthread_mutex_init(&log_mutex, 0);
2934 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2937 * XXX: The assumptions below are a bit naive
2938 * and can cause locks that aren't necessary.
2940 * For example older FreeBSD versions (< 6.x?)
2941 * have no gethostbyname_r, but gethostbyname is
2944 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2945 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2946 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2948 * XXX: should we use a single mutex for
2949 * localtime() and gmtime() as well?
2951 #ifndef HAVE_GMTIME_R
2952 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2953 #endif /* ndef HAVE_GMTIME_R */
2955 #ifndef HAVE_LOCALTIME_R
2956 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2957 #endif /* ndef HAVE_GMTIME_R */
2960 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2961 #endif /* ndef HAVE_RANDOM */
2962 #endif /* FEATURE_PTHREAD */
2965 * TODO: mutex support for mingw32 would be swell.
2970 printf("Fatal error. Mutex initialization failed: %s.\n",
2979 /*********************************************************************
2983 * Description : Load the config file and start the listen loop.
2984 * This function is a lot more *sane* with the `load_config'
2985 * and `listen_loop' functions; although it stills does
2986 * a *little* too much for my taste.
2989 * 1 : argc = Number of parameters (including $0).
2990 * 2 : argv = Array of (char *)'s to the parameters.
2992 * Returns : 1 if : can't open config file, unrecognized directive,
2993 * stats requested in multi-thread mode, can't open the
2994 * log file, can't open the jar file, listen port is invalid,
2995 * any load fails, and can't bind port.
2997 * Else main never returns, the process must be signaled
2998 * to terminate execution. Or, on Windows, use the
2999 * "File", "Exit" menu option.
3001 *********************************************************************/
3003 int real_main(int argc, const char *argv[])
3005 int main(int argc, const char *argv[])
3009 unsigned int random_seed;
3011 struct passwd *pw = NULL;
3012 struct group *grp = NULL;
3015 char *pre_chroot_nslookup_to_load_resolver = NULL;
3022 #if !defined(_WIN32)
3030 * Parse the command line arguments
3032 * XXX: simply printing usage information in case of
3033 * invalid arguments isn't particularly user friendly.
3035 while (++argc_pos < argc)
3038 /* Check to see if the service must be installed or uninstalled */
3039 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3041 const char *pName = argv[argc_pos] + 9;
3044 exit( (install_service(pName)) ? 0 : 1 );
3046 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3048 const char *pName = argv[argc_pos] + 11;
3051 exit((uninstall_service(pName)) ? 0 : 1);
3053 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3055 bRunAsService = TRUE;
3056 w32_set_service_cwd();
3057 atexit(w32_service_exit_notify);
3060 #endif /* defined(_WIN32) */
3063 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3065 if (strcmp(argv[argc_pos], "--help") == 0)
3070 else if(strcmp(argv[argc_pos], "--version") == 0)
3072 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3078 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3083 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3085 if (++argc_pos == argc) usage(argv[0]);
3086 pidfile = strdup(argv[argc_pos]);
3089 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3091 if (++argc_pos == argc) usage(argv[argc_pos]);
3093 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3096 if (NULL == (grp = getgrnam(p)))
3098 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3102 if (NULL == (pw = getpwnam(argv[argc_pos])))
3104 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3107 if (p != NULL) *--p = '\0';
3110 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3112 if (++argc_pos == argc) usage(argv[0]);
3113 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3116 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3120 #endif /* defined(unix) */
3122 else if (argc_pos + 1 != argc)
3125 * This is neither the last command line
3126 * option, nor was it recognized before,
3127 * therefore it must be invalid.
3133 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3135 configfile = argv[argc_pos];
3138 } /* -END- while (more arguments) */
3141 if ( *configfile != '/' )
3143 char cwd[BUFFER_SIZE];
3145 size_t abs_file_size;
3147 /* make config-filename absolute here */
3148 if (NULL == getcwd(cwd, sizeof(cwd)))
3150 perror("failed to get current working directory");
3155 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3156 basedir = strdup(cwd);
3158 if (NULL == basedir ||
3159 NULL == (abs_file = malloc(abs_file_size)))
3161 perror("malloc failed");
3164 strlcpy(abs_file, basedir, abs_file_size);
3165 strlcat(abs_file, "/", abs_file_size );
3166 strlcat(abs_file, configfile, abs_file_size);
3167 configfile = abs_file;
3169 #endif /* defined unix */
3173 clients->next = NULL;
3175 /* XXX: factor out initialising after the next stable release. */
3178 #elif defined(_WIN32)
3182 /* Prepare mutexes if supported and necessary. */
3183 initialize_mutexes();
3185 /* Enable logging until further notice. */
3186 init_log_module(Argv[0]);
3188 random_seed = (unsigned int)time(NULL);
3190 srandom(random_seed);
3191 #elif defined (_WIN32)
3193 * See pick_from_range() in miscutil.c for details.
3195 log_error(LOG_LEVEL_INFO,
3196 "No thread-safe PRNG implemented for your platform. "
3197 "Using weak \'randomization\' factor which will "
3198 "limit the already questionable usefulness of "
3199 "header-time-randomizing actions (disabled by default).");
3202 #endif /* ifdef HAVE_RANDOM */
3205 * Unix signal handling
3207 * Catch the abort, interrupt and terminate signals for a graceful exit
3208 * Catch the hangup signal so the errlog can be reopened.
3209 * Ignore the broken pipe signals (FIXME: Why?)
3211 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3214 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3215 const int ignored_signals[] = { SIGPIPE, 0 };
3217 for (idx = 0; catched_signals[idx] != 0; idx++)
3219 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3220 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3222 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3223 #endif /* ifdef sun */
3225 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3229 for (idx = 0; ignored_signals[idx] != 0; idx++)
3231 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3233 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3238 #else /* ifdef _WIN32 */
3239 # ifdef _WIN_CONSOLE
3241 * We *are* in a windows console app.
3242 * Print a verbose messages about FAQ's and such
3244 printf("%s", win32_blurb);
3245 # endif /* def _WIN_CONSOLE */
3246 #endif /* def _WIN32 */
3249 /* Initialize the CGI subsystem */
3250 cgi_init_error_messages();
3253 * If runnig on unix and without the --nodaemon
3254 * option, become a daemon. I.e. fork, detach
3255 * from tty and get process group leadership
3268 if ( pid < 0 ) /* error */
3273 else if ( pid != 0 ) /* parent */
3278 * must check for errors
3279 * child died due to missing files aso
3282 wpid = waitpid( pid, &status, WNOHANG );
3291 /* Should be more portable, but not as well tested */
3296 #else /* ndef __FreeBSD__ */
3298 #endif /* ndef __FreeBSD__ */
3299 fd = open("/dev/tty", O_RDONLY);
3302 /* no error check here */
3303 ioctl( fd, TIOCNOTTY,0 );
3308 * stderr (fd 2) will be closed later on, when the
3309 * log file has been parsed.
3316 } /* -END- if (!no_daemon) */
3319 * As soon as we have written the PID file, we can switch
3320 * to the user and group ID indicated by the --user option
3326 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3328 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3334 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3336 /* Read the time zone file from /etc before doing chroot. */
3338 if (NULL != pre_chroot_nslookup_to_load_resolver
3339 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3341 /* Initialize resolver library. */
3342 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3344 if (chroot(pw->pw_dir) < 0)
3346 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3350 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3353 if (setuid(pw->pw_uid))
3355 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3359 char putenv_dummy[64];
3361 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3362 if (putenv(putenv_dummy) != 0)
3364 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3367 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3368 if (putenv(putenv_dummy) != 0)
3370 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3376 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3379 #endif /* defined unix */
3382 /* This will be FALSE unless the command line specified --service
3386 /* Yup, so now we must attempt to establish a connection
3387 * with the service dispatcher. This will only work if this
3388 * process was launched by the service control manager to
3389 * actually run as a service. If this isn't the case, i've
3390 * known it take around 30 seconds or so for the call to return.
3393 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3394 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3396 /* Service has run, and at this point is now being stopped, so just return */
3401 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3403 /* An error occurred. Usually it's because --service was wrongly specified
3404 * and we were unable to connect to the Service Control Dispatcher because
3405 * it wasn't expecting us and is therefore not listening.
3407 * For now, just continue below to call the listen_loop function.
3410 #endif /* def _WIN32 */
3420 /*********************************************************************
3422 * Function : bind_port_helper
3424 * Description : Bind the listen port. Handles logging, and aborts
3428 * 1 : config = Privoxy configuration. Specifies port
3431 * Returns : Port that was opened.
3433 *********************************************************************/
3434 static jb_socket bind_port_helper(struct configuration_spec * config)
3439 if (config->haddr == NULL)
3441 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3446 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3447 config->hport, config->haddr);
3450 result = bind_port(config->haddr, config->hport, &bfd);
3457 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3458 "There may be another Privoxy or some other "
3459 "proxy running on port %d",
3460 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3461 config->hport, config->hport);
3464 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3465 "The hostname is not resolvable",
3466 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3469 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3470 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3473 /* shouldn't get here */
3474 return JB_INVALID_SOCKET;
3477 config->need_bind = 0;
3484 /* Without this simple workaround we get this compiler warning from _beginthread
3485 * warning C4028: formal parameter 1 different from declaration
3487 void w32_service_listen_loop(void *p)
3491 #endif /* def _WIN32 */
3494 /*********************************************************************
3496 * Function : listen_loop
3498 * Description : bind the listen port and enter a "FOREVER" listening loop.
3504 *********************************************************************/
3505 static void listen_loop(void)
3507 struct client_state *csp = NULL;
3509 struct configuration_spec * config;
3511 config = load_config();
3513 bfd = bind_port_helper(config);
3515 #ifdef FEATURE_GRACEFUL_TERMINATION
3516 while (!g_terminate)
3521 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3522 while (waitpid(-1, NULL, WNOHANG) > 0)
3524 /* zombie children */
3526 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3529 * Free data that was used by died threads
3535 * Re-open the errlog after HUP signal
3537 if (received_hup_signal)
3539 init_error_log(Argv[0], config->logfile);
3540 received_hup_signal = 0;
3545 #ifdef FEATURE_COOKIE_JAR
3547 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3548 * already open. I think unload_configfile() is not being run, which should do
3549 * this work. Until that can get resolved, we'll use this workaround.
3553 if (csp->config->jar)
3555 fclose(csp->config->jar);
3556 csp->config->jar = NULL;
3558 #endif /* FEATURE_COOKIE_JAR */
3559 #endif /* __OS2__ */
3561 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3563 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3567 csp->flags |= CSP_FLAG_ACTIVE;
3568 csp->sfd = JB_INVALID_SOCKET;
3570 csp->config = config = load_config();
3572 if ( config->need_bind )
3575 * Since we were listening to the "old port", we will not see
3576 * a "listen" param change until the next IJB request. So, at
3577 * least 1 more request must be made for us to find the new
3578 * setting. I am simply closing the old socket and binding the
3581 * Which-ever is correct, we will serve 1 more page via the
3582 * old settings. This should probably be a "show-proxy-args"
3583 * request. This should not be a so common of an operation
3584 * that this will hurt people's feelings.
3589 bfd = bind_port_helper(config);
3592 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3594 if (!accept_connection(csp, bfd))
3596 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3609 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3612 #ifdef FEATURE_TOGGLE
3613 if (global_toggle_state)
3614 #endif /* def FEATURE_TOGGLE */
3616 csp->flags |= CSP_FLAG_TOGGLED_ON;
3619 if (run_loader(csp))
3621 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3622 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3626 if (block_acl(NULL,csp))
3628 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3629 close_socket(csp->cfd);
3633 #endif /* def FEATURE_ACL */
3635 /* add it to the list of clients */
3636 csp->next = clients->next;
3637 clients->next = csp;
3639 if (config->multi_threaded)
3643 /* this is a switch () statment in the C preprocessor - ugh */
3644 #undef SELECTED_ONE_OPTION
3646 /* Use Pthreads in preference to native code */
3647 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3648 #define SELECTED_ONE_OPTION
3650 pthread_t the_thread;
3651 pthread_attr_t attrs;
3653 pthread_attr_init(&attrs);
3654 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3655 errno = pthread_create(&the_thread, &attrs,
3656 (void * (*)(void *))serve, csp);
3657 child_id = errno ? -1 : 0;
3658 pthread_attr_destroy(&attrs);
3662 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3663 #define SELECTED_ONE_OPTION
3664 child_id = _beginthread(
3665 (void (*)(void *))serve,
3670 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3671 #define SELECTED_ONE_OPTION
3672 child_id = _beginthread(
3673 (void(* _Optlink)(void*))serve,
3679 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3680 #define SELECTED_ONE_OPTION
3682 thread_id tid = spawn_thread
3683 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3685 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3687 child_id = (int) tid;
3696 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3697 #define SELECTED_ONE_OPTION
3698 csp->cfd = ReleaseSocket(csp->cfd, -1);
3701 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3702 NP_Output, Output(),
3703 NP_CloseOutput, FALSE,
3704 NP_Name, (ULONG)"privoxy child",
3708 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3709 NP_Output, Output(),
3710 NP_CloseOutput, FALSE,
3711 NP_Name, (ULONG)"privoxy child",
3712 NP_StackSize, 200*1024,
3718 ((struct Task *)child_id)->tc_UserData = csp;
3719 Signal((struct Task *)child_id, SIGF_SINGLE);
3724 #if !defined(SELECTED_ONE_OPTION)
3727 /* This block is only needed when using fork().
3728 * When using threads, the server thread was
3729 * created and run by the call to _beginthread().
3731 if (child_id == 0) /* child */
3734 #ifdef FEATURE_TOGGLE
3735 int inherited_toggle_state = global_toggle_state;
3736 #endif /* def FEATURE_TOGGLE */
3741 * If we've been toggled or we've blocked the request, tell Mom
3744 #ifdef FEATURE_TOGGLE
3745 if (inherited_toggle_state != global_toggle_state)
3747 rc |= RC_FLAG_TOGGLED;
3749 #endif /* def FEATURE_TOGGLE */
3751 #ifdef FEATURE_STATISTICS
3752 if (csp->flags & CSP_FLAG_REJECTED)
3754 rc |= RC_FLAG_BLOCKED;
3756 #endif /* ndef FEATURE_STATISTICS */
3760 else if (child_id > 0) /* parent */
3762 /* in a fork()'d environment, the parent's
3763 * copy of the client socket and the CSP
3767 #if !defined(_WIN32) && !defined(__CYGWIN__)
3769 wait( &child_status );
3772 * Evaluate child's return code: If the child has
3773 * - been toggled, toggle ourselves
3774 * - blocked its request, bump up the stats counter
3777 #ifdef FEATURE_TOGGLE
3778 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3780 global_toggle_state = !global_toggle_state;
3782 #endif /* def FEATURE_TOGGLE */
3784 #ifdef FEATURE_STATISTICS
3786 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3790 #endif /* def FEATURE_STATISTICS */
3792 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3793 close_socket(csp->cfd);
3794 csp->flags &= ~CSP_FLAG_ACTIVE;
3798 #undef SELECTED_ONE_OPTION
3799 /* end of cpp switch () */
3801 if (child_id < 0) /* failed */
3803 char buf[BUFFER_SIZE];
3805 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3807 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3809 write_socket(csp->cfd, buf, strlen(buf));
3810 close_socket(csp->cfd);
3811 csp->flags &= ~CSP_FLAG_ACTIVE;
3822 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3824 /* Clean up. Aim: free all memory (no leaks) */
3825 #ifdef FEATURE_GRACEFUL_TERMINATION
3827 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3829 unload_current_config_file();
3830 unload_current_actions_file();
3831 unload_current_re_filterfile();
3832 #ifdef FEATURE_TRUST
3833 unload_current_trust_file();
3836 if (config->multi_threaded)
3843 } while ((clients->next != NULL) && (--i > 0));
3847 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3858 #ifdef FEATURE_COOKIE_JAR
3859 if (NULL != config->jar)
3861 fclose(config->jar);
3865 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3866 /* Cleanup - remove taskbar icon etc. */
3871 #endif /* FEATURE_GRACEFUL_TERMINATION */