1 const char jcc_rcs[] = "$Id: jcc.c,v 1.172 2008/04/16 16:38:21 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 1.172 2008/04/16 16:38:21 fabiankeil
37 * Don't pass the whole csp structure to flush_socket()
38 * when it only needs a file descriptor and a buffer.
40 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
41 * Remove kill-popups action.
43 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
44 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
46 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
47 * Remove the treat-forbidden-connects-like-blocks action. We now
48 * use the "blocked" page for forbidden CONNECT requests by default.
50 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
51 * Also use shiny new connect_port_is_forbidden() in jcc.c.
53 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
54 * Rename url_actions() to get_url_actions() and let it
55 * use the standard parameter ordering.
57 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
58 * Let forward_url() use the standard parameter ordering
59 * and mark its second parameter immutable.
61 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
62 * Remove the "Listening ... for local connections only" log message.
63 * Whether or not remote connections are able to reach Privoxy is up
64 * to the operating system.
66 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
67 * Prevent the log messages for CONNECT requests to unacceptable
68 * ports from printing the limit-connect argument as [null] if
69 * limit-connect hasn't been explicitly enabled.
71 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
72 * Make sure all console-mode apps get a usage() instance
74 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
75 * Reword NO_SERVER_DATA_RESPONSE to make it harder
76 * to misunderstand what the message is all about.
78 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
79 * Unbreak trustfile which previously didn't work without
80 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
82 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
83 * Plug memory leak. Spotted by Valgrind, triggered by
84 * Privoxy-Regression-Test feeding proxyfuzz.py.
86 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
87 * In the HTTP snipplets, refer to the client as client.
89 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
90 * Emit a log message when activating the MS IIS5 hack.
92 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
93 * Log the "weak randomization factor" warning only
94 * once for mingw32 and provide some more details.
96 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
97 * Initialize log module after initializing mutexes, future
98 * deadlocks in that code should now work cross-platform.
100 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
101 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
102 * Reported by Bert van Leeuwen in BR#1818808.
104 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
105 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
107 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
108 * When in daemon mode, close stderr after the configuration file has been
109 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
111 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
112 * - Fix a crash when parsing invalid requests whose first header
113 * is rejected by get_header(). Regression (re?)introduced
114 * in r1.143 by yours truly.
115 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
116 * client_cookie_adder() to make sure send-vanilla-wafer can be
117 * controlled through tags (and thus regression-tested).
119 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
120 * - Move get_filter_function() from jcc.c to filters.c
121 * so the filter functions can be static.
122 * - Don't bother filtering body-less responses.
124 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
125 * Execute content filters through execute_content_filter().
127 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
128 * Initialize req to NULL to make sure it's defined if the
129 * first read_socket() call fails. Reported by icmp30.
131 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
132 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
133 * extensive comments moved to user manual.
135 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
136 * Don't crash if a broken header filter wiped out the request line.
138 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
139 * Fix byte_count calculation in case of flushes
140 * and don't parse the server headers a second time.
142 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
143 * - If there's a connection problem after we already forwarded
144 * parts of the original content, just hang up. Fixes BR#1776724.
145 * - Fix warnings about unused code on mingw32.
146 * - In case of flushes, calculate the byte count
147 * less incorrectly (I think).
149 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
150 * Add some more prototypes for static functions.
152 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
153 * Comment out request_contains_null_bytes() until it's used again.
155 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
156 * #1763173 from Stefan Huehner: s@const static@static const@
157 * and declare some more functions static.
159 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
160 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
161 * and explain why they were rejected in the first place.
162 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
163 * CONNECT requests. The request line was missing.
164 * - Add two more XXX reminders as we don't have enough already.
166 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
167 * As Hal noticed, checking dispatch_cgi() as the last cruncher
168 * looks like a bug if CGI requests are blocked unintentionally,
169 * so don't do it unless the user enabled the new config option
170 * "allow-cgi-request-crunching".
172 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
173 * - Allow to rewrite the request destination behind the client's back.
174 * - Turn the weird-looking unconditional for loop that
175 * reads the client request into a conditional while loop.
176 * Move the stuff that only runs once out of the loop.
177 * - Move parts of chat(), server_content_type() and the
178 * necessary stuff to fix BR#1750917 into get_filter_function().
180 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
181 * Temporary workaround for BR#1730105.
183 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
184 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
185 * deadlocks and crashes on OpenBSD and possibly other OS with neither
186 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
187 * Thanks to Ralf Horstmann for report and solution.
189 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
190 * Add forward-override{} to change the forwarding settings through
191 * action sections. This is mainly interesting to forward different
192 * clients differently (for example based on User-Agent or request
195 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
196 * - Let usage() mention the --chroot parameter.
197 * - Use read_socket() consistently and always leave
198 * the last buffer byte alone, even in cases where
199 * null termination (currently) doesn't matter.
201 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
202 * - Fix config file loading on Unix if no config file is specified.
203 * Since r1.97 Privoxy would always interpret the last argument as
204 * config file, even if it's a valid command line option.
205 * - Abort in case of unrecognized command line options. Closes #1719696.
206 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
207 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
209 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
210 * - Don't rerun crunchers that only depend on the request URL.
211 * - Don't count redirects and CGI requests as "blocked requests".
213 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
214 * Support crunching based on tags created by server-header taggers.
216 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
217 * Make HTTP snippets static (again). Add a Content-Type for those
218 * with content so the browser doesn't guess it based on the URL.
220 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
221 * Move crunching and request line rebuilding out of chat().
223 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
224 * Introduce tags as alternative way to specify which
225 * actions apply to a request. At the moment tags can be
226 * created based on client and server headers.
228 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
229 * Don't CLF-log CONNECT requests twice.
231 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
232 * Log the source address for ACL-related connection drops.
234 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
235 * New config option: enforce-blocks.
237 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
238 * - Move null byte check into separate function.
239 * - Don't confuse the client with error pages
240 * if a CONNECT request was already confirmed.
242 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
243 * Speed up NULL byte escaping and only log the complete
244 * NULL byte requests with header debugging enabled.
246 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
247 * Answer requests that contain NULL bytes with
248 * a custom response instead of waiting for more
249 * data until the client eventually hangs up.
251 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
252 * - Move delivery and logging of crunched responses
253 * from chat() into send_crunch_response().
254 * - Display the reason for generating http_responses.
255 * - Log the content length for LOG_LEVEL_CLF correctly
256 * (still incorrect for some fixed responses).
257 * - Reword an incorrect comment about
258 * treat-forbidden-connects-like-blocks violating
260 * - Add some log messages.
262 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
263 * Move mutex initialization into separate
264 * function and exit in case of errors.
266 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
267 * - Start to reduce chat()'s line count and move
268 * parts of it into separate functions.
269 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
271 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
272 * - Add Proxy-Agent header to HTTP snippets that are
273 * supposed to reach HTTP clients only.
274 * - Made a few CONNECT log messages more descriptive.
275 * - Catch completely empty server responses (as seen
276 * with Tor's fake ".noconnect" top level domain).
277 * - Use shiny new "forwarding-failed" template for socks errors.
279 * Revision 1.118 2007/01/07 07:43:43 joergs
280 * AmigaOS4 support added.
282 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
283 * Added config option accept-intercepted-requests
284 * and disabled it by default.
286 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
287 * Reverted parts of my last commit
288 * to keep error handling working.
290 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
291 * Fixed gcc43 conversion warnings.
293 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
294 * Fix -pedantic ISO C warning about converting
295 * from function pointer to object pointer.
297 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
298 * Silence compiler warning I introduced with my last commit.
300 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
301 * Mutex protect rand() if POSIX threading
302 * is used, warn the user if that's not possible
303 * and stop using it on _WIN32 where it could
306 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
307 * Don't prevent core dumps by catching SIGABRT.
308 * It's rude and makes debugging unreasonable painful.
310 * Revision 1.110 2006/12/13 14:52:53 etresoft
311 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
313 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
314 * Privoxy is now able to run as intercepting
315 * proxy in combination with any packet filter
316 * that does the port redirection. The destination
317 * is extracted from the "Host:" header which
318 * should be available for nearly all requests.
320 * Moved HTTP snipplets into jcc.c.
321 * Added error message for gopher proxy requests.
323 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
324 * Only unlink the pidfile if it's actually used.
326 * Change order of interception checks to make
327 * it possible to block or redirect requests for
330 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
331 * Make pthread mutex locking more generic. Instead of
332 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
333 * and use mutex locking unless there is an _r function
334 * available. Better safe than sorry.
336 * Fixes "./configure --disable-pthread" and should result
337 * in less threading-related problems on pthread-using platforms,
338 * but it still doesn't fix BR#1122404.
340 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
341 * Move pthread.h inclusion from jcc.c to jcc.h.
342 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
344 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
345 * Don't exit after receiving the second SIGHUP on Solaris.
347 * Fixes BR 1052235, but the same problem may exist on other
348 * systems. Once 3.0.6 is out we should use sigset()
349 * where available and see if it breaks anything.
351 * Revision 1.104 2006/09/23 13:26:38 roro
352 * Replace TABs by spaces in source code.
354 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
355 * Fix +redirect{}. Didn't work with -fast-redirects.
357 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
358 * Respond with 400 and a short text message
359 * if the client tries to use Privoxy as FTP proxy.
361 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
362 * Make number of retries in case of forwarded-connect problems
363 * a config file option (forwarded-connect-retries) and use 0 as
366 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
367 * Set random(3) seed.
369 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
370 * Follow the OpenBSD port's lead and protect the resolve
371 * functions on OpenBSD as well.
373 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
374 * --user fix. Only use the user as group if no group is specified.
375 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
377 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
378 * Windows service (re-)integration
382 * --install[:service_name]
383 * --uninstall[:service_name]
386 * They work as follows:
387 * --install will create a service for you and then terminate.
388 * By default the service name will be "privoxy" (without the quotes).
389 * However you can run multiple services if you wish, just by adding
390 * a colon and then a name (no spaces).
392 * --uninstall follows the exact same rules a --install.
394 * --service is used when the program is executed by the service
395 * control manager, and in normal circumstances would never be
396 * used as a command line argument.
398 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
399 * Windows service integration
401 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
402 * Incorporate Fabian Keil's patch work:
403 http://www.fabiankeil.de/sourcecode/privoxy/
405 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
406 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
407 * with what was really the latest development (the v_3_0_branch branch)
409 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
410 * Thanks to Jindrich Makovicka for a race condition fix for the log
411 * file. The race condition remains for non-pthread implementations.
412 * Reference patch #1175720.
414 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
415 * Add the ability to check jpeg images for invalid
416 * lengths of comment blocks. Defensive strategy
417 * against the exploit:
418 * Microsoft Security Bulletin MS04-028
419 * Buffer Overrun in JPEG Processing (GDI+) Could
420 * Allow Code Execution (833987)
421 * Enabled with +inspect-jpegs in actions files.
423 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
424 * - Fixed usage info for non-unix platforms
425 * - Fixed small cmdline parsing bug
427 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
428 * Diagnostics: Now preserve the returncode of pthread_create
429 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
431 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
432 * No longer ignore SIGCHLD. Fixes bug #769381
434 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
435 * Close jarfile on graceful exit, remove stray line
437 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
438 * Cosmetics: Killed a warning, a typo and an allocation left at exit
440 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
441 * No longer rely on non-POSIX.1 extensions of getcwd().
444 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
445 * Replaced setenv() by posix-compliant putenv()
446 * Thanks to Neil McCalden (nmcc AT users.sf.net).
448 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
449 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
451 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
452 * Clean-up and extension of improvements for forked mode:
453 * - Child's return code now consists of flags RC_FLAG_*
454 * - Reporting toggle to parent now properly #ifdef'ed
455 * - Children now report blocking to parent. This enables
456 * statistics in forked mode
458 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
459 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
460 * Children now report having been toggled through _exit(17), parents
461 * watch for that code and toggle themselves if found.
463 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
464 * Wrapping all *_r functions (the non-_r versions of them) with
465 * mutex semaphores for OSX. Hopefully this will take care of all
466 * of those pesky crash reports.
468 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
469 * Fixed two mostly harmless mem leaks
471 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
472 * Fix: Head of global clients list now initialized to NULL
474 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
475 * Added basic support for OPTIONS and TRACE HTTP methods:
476 * - New interceptor direct_response() added in chat().
477 * - sed() moved to earlier in the process, so that the
478 * Host: header is evaluated before actions and forwarding
481 * Revision 1.92 2002/05/08 16:00:46 oes
482 * Chat's buffer handling:
483 * - Fixed bug with unchecked out-of-mem conditions
484 * while reading client request & server headers
485 * - No longer predict if the buffer limit will be exceeded
486 * in the next read -- check add_to_iob's new
487 * return code. If buffer couldn't be extended
488 * (policy or out-of-mem) while
489 * - reading from client: abort
490 * - reading server headers: send error page
491 * - buffering server body for filter: flush,
492 * and if that fails: send error page
494 * Revision 1.91 2002/04/08 20:35:58 swa
497 * Revision 1.90 2002/04/02 14:57:28 oes
498 * Made sending wafers independent of FEATURE_COOKIE_JAR
500 * Revision 1.89 2002/03/31 17:18:59 jongfoster
501 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
503 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
504 * More compiler warning message maintenance
506 * Revision 1.87 2002/03/26 22:29:54 swa
507 * we have a new homepage!
509 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
510 * Workaround for closing the jarfile before load_config() comes around again
512 * Revision 1.85 2002/03/24 15:23:33 jongfoster
515 * Revision 1.84 2002/03/24 13:25:43 swa
516 * name change related issues
518 * Revision 1.83 2002/03/16 23:54:06 jongfoster
519 * Adding graceful termination feature, to help look for memory leaks.
520 * If you enable this (which, by design, has to be done by hand
521 * editing config.h) and then go to http://i.j.b/die, then the program
522 * will exit cleanly after the *next* request. It should free all the
523 * memory that was used.
525 * Revision 1.82 2002/03/13 00:27:05 jongfoster
528 * Revision 1.81 2002/03/12 01:42:50 oes
529 * Introduced modular filters
531 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
532 * OS/2 port maintenance:
533 * - Fixed EMX build - it had decayed a little
534 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
535 * substituted a memset for now.
537 * Revision 1.79 2002/03/09 20:03:52 jongfoster
538 * - Making various functions return int rather than size_t.
539 * (Undoing a recent change). Since size_t is unsigned on
540 * Windows, functions like read_socket that return -1 on
541 * error cannot return a size_t.
543 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
544 * crashes, and also frequently caused JB to jump to 100%
545 * CPU and stay there. (Because it thought it had just
546 * read ((unsigned)-1) == 4Gb of data...)
548 * - The signature of write_socket has changed, it now simply
549 * returns success=0/failure=nonzero.
551 * - Trying to get rid of a few warnings --with-debug on
552 * Windows, I've introduced a new type "jb_socket". This is
553 * used for the socket file descriptors. On Windows, this
554 * is SOCKET (a typedef for unsigned). Everywhere else, it's
555 * an int. The error value can't be -1 any more, so it's
556 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
557 * Windows it maps to the #define INVALID_SOCKET.)
559 * - The signature of bind_port has changed.
561 * Revision 1.78 2002/03/08 21:35:04 oes
562 * Added optional group supplement to --user option. Will now use default group of user if no group given
564 * Revision 1.77 2002/03/07 03:52:06 oes
565 * - Fixed compiler warnings etc
566 * - Improved handling of failed DNS lookups
568 * Revision 1.76 2002/03/06 22:54:35 jongfoster
569 * Automated function-comment nitpicking.
571 * Revision 1.75 2002/03/06 10:02:19 oes
572 * Fixed stupid bug when --user was not given
574 * Revision 1.74 2002/03/06 00:49:31 jongfoster
575 * Fixing warning on Windows
576 * Making #ifdefs that refer to the same variable consistently
577 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
579 * Revision 1.73 2002/03/05 23:57:30 hal9
580 * Stray character 's' on line 1618 was breaking build.
582 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
583 * - Re-enable OS/2 building after new parms were added
584 * - Fix false out of memory report when resolving CGI templates when no IP
585 * address is available of failed attempt (a la no such domain)
587 * Revision 1.71 2002/03/05 18:13:56 oes
588 * Added --user option
590 * Revision 1.70 2002/03/05 04:52:42 oes
591 * Deleted non-errlog debugging code
593 * Revision 1.69 2002/03/04 23:50:00 jongfoster
594 * Splitting off bind_port() call into bind_port_helper(), with
597 * Revision 1.68 2002/03/04 20:17:32 oes
600 * Revision 1.67 2002/03/04 18:18:57 oes
601 * - Removed _DEBUG mode
602 * - Cleand up cmdline parsing
603 * - Introduced --no-daemon, --pidfile options
604 * - Cleaned up signal handling:
605 * - Terminate cleanly on INT, TERM and ABRT
606 * - Schedule logfile for re-opening on HUP
607 * - Ignore CHLD and PIPE
608 * - Leave the rest with their default handlers
609 * - Uniform handler registration
610 * - Added usage() function
611 * - Played styleguide police
613 * Revision 1.66 2002/03/03 15:06:55 oes
614 * Re-enabled automatic config reloading
616 * Revision 1.65 2002/03/03 14:49:11 oes
617 * Fixed CLF logging: Now uses client's original HTTP request
619 * Revision 1.64 2002/03/03 09:18:03 joergs
620 * Made jumbjuster work on AmigaOS again.
622 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
623 * Clean up a little CRLF unpleasantness that suddenly appeared
625 * Revision 1.62 2002/02/20 23:17:23 jongfoster
626 * Detecting some out-of memory conditions and exiting with a log message.
628 * Revision 1.61 2002/01/17 21:01:52 jongfoster
629 * Moving all our URL and URL pattern parsing code to urlmatch.c.
631 * Revision 1.60 2001/12/30 14:07:32 steudten
632 * - Add signal handling (unix)
633 * - Add SIGHUP handler (unix)
634 * - Add creation of pidfile (unix)
635 * - Add action 'top' in rc file (RH)
636 * - Add entry 'SIGNALS' to manpage
637 * - Add exit message to logfile (unix)
639 * Revision 1.59 2001/12/13 14:07:18 oes
640 * Fixed Bug: 503 error page now sent OK
642 * Revision 1.58 2001/11/30 23:37:24 jongfoster
643 * Renaming the Win32 config file to config.txt - this is almost the
644 * same as the corresponding UNIX name "config"
646 * Revision 1.57 2001/11/16 00:47:43 jongfoster
647 * Changing the tty-disconnection code to use setsid().
649 * Revision 1.56 2001/11/13 20:20:54 jongfoster
650 * Tabs->spaces, fixing a bug with missing {} around an if()
652 * Revision 1.55 2001/11/13 20:14:53 jongfoster
653 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
655 * Revision 1.54 2001/11/07 00:03:14 steudten
656 * Give reliable return value if an error
657 * occurs not just 0 with new daemon mode.
659 * Revision 1.53 2001/11/05 21:41:43 steudten
660 * Add changes to be a real daemon just for unix os.
661 * (change cwd to /, detach from controlling tty, set
662 * process group and session leader to the own process.
664 * Add some fatal-error log message for failed malloc().
665 * Add '-d' if compiled with 'configure --with-debug' to
666 * enable debug output.
668 * Revision 1.52 2001/10/26 20:11:20 jongfoster
669 * Fixing type mismatch
671 * Revision 1.51 2001/10/26 17:38:28 oes
674 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
675 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
676 * threads to call select() simultaneously. So, it's time to do a real, live,
677 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
678 * (native). Both versions will work, but using __OS2__ offers multi-threading.
680 * Revision 1.49 2001/10/23 21:41:35 jongfoster
681 * Added call to initialize the (statically-allocated of course)
682 * "out of memory" CGI response.
684 * Revision 1.48 2001/10/10 19:56:46 jongfoster
685 * Moving some code that wasn't cookie-related out of an #ifdef
688 * Revision 1.47 2001/10/10 16:44:36 oes
689 * Added CONNECT destination port limitation check
691 * Revision 1.46 2001/10/08 15:17:41 oes
692 * Re-enabled SSL forwarding
694 * Revision 1.45 2001/10/07 15:42:11 oes
695 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
697 * Moved downgrading of the HTTP version from parse_http_request to
698 * chat(), since we can't decide if it is necessary before we have
699 * determined the actions for the URL. The HTTP command is now
700 * *always* re-built so the repairs need no longer be special-cased.
702 * filter_popups now gets a csp pointer so it can raise the new
703 * CSP_FLAG_MODIFIED flag.
707 * Added configurable size limit for the IOB. If the IOB grows so
708 * large that the next read would exceed the limit, the header
709 * is generated, and the header & unfiltered buffer are flushed
710 * to the client. Chat then continues in non-buffering,
711 * non-filtering body mode.
713 * Revision 1.44 2001/10/02 18:13:57 oes
716 * Revision 1.43 2001/10/02 15:32:13 oes
717 * Moved generation of hdr
719 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
720 * Cleaning up 2 compiler warnings on OS/2.
722 * Revision 1.41 2001/09/16 17:05:14 jongfoster
723 * Removing unused #include showarg.h
725 * Revision 1.40 2001/09/16 15:41:45 jongfoster
726 * Fixing signed/unsigned comparison warning.
728 * Revision 1.39 2001/09/16 13:21:27 jongfoster
729 * Changes to use new list functions.
731 * Revision 1.38 2001/09/16 13:01:46 jongfoster
732 * Removing redundant function call that zeroed zalloc()'d memory.
734 * Revision 1.37 2001/09/10 11:12:24 oes
735 * Deleted unused variable
737 * Revision 1.36 2001/09/10 10:56:15 oes
738 * Silenced compiler warnings
740 * Revision 1.35 2001/07/31 14:44:22 oes
741 * Deleted unused size parameter from filter_popups()
743 * Revision 1.34 2001/07/30 22:08:36 jongfoster
744 * Tidying up #defines:
745 * - All feature #defines are now of the form FEATURE_xxx
746 * - Permanently turned off WIN_GUI_EDIT
747 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
749 * Revision 1.33 2001/07/29 19:32:00 jongfoster
750 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
752 * Revision 1.32 2001/07/29 18:47:05 jongfoster
753 * Adding missing #include "loadcfg.h"
755 * Revision 1.31 2001/07/29 12:17:48 oes
756 * Applied pthread fix by Paul Lieverse
758 * Revision 1.30 2001/07/25 22:57:13 jongfoster
759 * __BEOS__ no longer overrides FEATURE_PTHREAD.
760 * This is because FEATURE_PTHREAD will soon be widely used, so I
761 * want to keep it simple.
763 * Revision 1.29 2001/07/24 12:47:06 oes
764 * Applied BeOS support update by Eugenia
766 * Revision 1.28 2001/07/23 13:26:12 oes
767 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
769 * Revision 1.27 2001/07/19 19:09:47 haroon
770 * - Added code to take care of the situation where while processing the first
771 * server response (which includes the server header), after finding the end
772 * of the headers we were not looking past the end of the headers for
773 * content modification. I enabled it for filter_popups.
774 * Someone else should look to see if other similar operations should be
775 * done to the discarded portion of the buffer.
777 * Note 2001/07/20: No, the other content modification mechanisms will process
778 * the whole iob later anyway. --oes
780 * Revision 1.26 2001/07/18 12:31:36 oes
783 * Revision 1.25 2001/07/15 19:43:49 jongfoster
784 * Supports POSIX threads.
785 * Also removed some unused #includes.
787 * Revision 1.24 2001/07/13 14:00:40 oes
788 * - Generic content modification scheme:
789 * Each feature has its own applicability flag that is set
790 * from csp->action->flags.
791 * Replaced the "filtering" int flag , by a function pointer
792 * "content_filter" to the function that will do the content
793 * modification. If it is != NULL, the document will be buffered
794 * and processed through *content_filter, which must set
795 * csp->content_length and return a modified copy of the body
796 * or return NULL (on failiure).
797 * - Changed csp->is_text to the more generic bitmap csp->content_type
798 * which can currently take the valued CT_TEXT or CT_GIF
800 * - Removed all #ifdef PCRS
802 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
803 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
804 * the `block_acl' call. This prevents a compilation error when the user
805 * does not wish to use the "ACL" feature.
807 * Revision 1.22 2001/06/29 21:45:41 oes
808 * Indentation, CRLF->LF, Tab-> Space
810 * Revision 1.21 2001/06/29 13:29:36 oes
811 * - Cleaned up, improved comments
812 * - Unified all possible interceptors (CGI,
813 * block, trust, fast_redirect) in one
814 * place, with one (CGI) answer generation
815 * mechansim. Much clearer now.
816 * - Removed the GIF image generation, which
817 * is now done in filters.c:block_url()
818 * - Made error conditions like domain lookup
819 * failiure or (various) problems while talking
820 * to the server use cgi.c:error_response()
821 * instead of generating HTML/HTTP in chat() (yuck!)
822 * - Removed logentry from cancelled commit
824 * Revision 1.20 2001/06/09 10:55:28 jongfoster
825 * Changing BUFSIZ ==> BUFFER_SIZE
827 * Revision 1.19 2001/06/07 23:12:52 jongfoster
828 * Replacing function pointer in struct gateway with a directly
829 * called function forwarded_connect().
830 * Replacing struct gateway with struct forward_spec
832 * Revision 1.18 2001/06/03 19:12:16 oes
833 * introduced new cgi handling
835 * Revision 1.17 2001/06/01 20:07:23 jongfoster
836 * Now uses action +image-blocker{} rather than config->tinygif
838 * Revision 1.16 2001/06/01 18:49:17 jongfoster
839 * Replaced "list_share" with "list" - the tiny memory gain was not
840 * worth the extra complexity.
842 * Revision 1.15 2001/05/31 21:24:47 jongfoster
843 * Changed "permission" to "action" throughout.
844 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
845 * Moved vanilla wafer check into chat(), since we must now
846 * decide whether or not to add it based on the URL.
848 * Revision 1.14 2001/05/29 20:14:01 joergs
849 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
852 * Revision 1.13 2001/05/29 09:50:24 jongfoster
853 * Unified blocklist/imagelist/permissionslist.
854 * File format is still under discussion, but the internal changes
857 * Also modified interceptor behaviour:
858 * - We now intercept all URLs beginning with one of the following
859 * prefixes (and *only* these prefixes):
861 * * http://ijbswa.sf.net/config/
862 * * http://ijbswa.sourceforge.net/config/
863 * - New interceptors "home page" - go to http://i.j.b/ to see it.
864 * - Internal changes so that intercepted and fast redirect pages
865 * are not replaced with an image.
866 * - Interceptors now have the option to send a binary page direct
867 * to the client. (i.e. ijb-send-banner uses this)
868 * - Implemented show-url-info interceptor. (Which is why I needed
869 * the above interceptors changes - a typical URL is
870 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
871 * The previous mechanism would not have intercepted that, and
872 * if it had been intercepted then it then it would have replaced
875 * Revision 1.12 2001/05/27 22:17:04 oes
877 * - re_process_buffer no longer writes the modified buffer
878 * to the client, which was very ugly. It now returns the
879 * buffer, which it is then written by chat.
881 * - content_length now adjusts the Content-Length: header
882 * for modified documents rather than crunch()ing it.
883 * (Length info in csp->content_length, which is 0 for
884 * unmodified documents)
886 * - For this to work, sed() is called twice when filtering.
888 * Revision 1.11 2001/05/26 17:27:53 jongfoster
889 * Added support for CLF and fixed LOG_LEVEL_LOG.
890 * Also did CRLF->LF fix of my previous patch.
892 * Revision 1.10 2001/05/26 15:26:15 jongfoster
893 * ACL feature now provides more security by immediately dropping
894 * connections from untrusted hosts.
896 * Revision 1.9 2001/05/26 00:28:36 jongfoster
897 * Automatic reloading of config file.
898 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
899 * Most of the global variables have been moved to a new
900 * struct configuration_spec, accessed through csp->config->globalname
901 * Most of the globals remaining are used by the Win32 GUI.
903 * Revision 1.8 2001/05/25 22:43:18 jongfoster
904 * Fixing minor memory leak and buffer overflow.
906 * Revision 1.7 2001/05/25 22:34:30 jongfoster
909 * Revision 1.6 2001/05/23 00:13:58 joergs
910 * AmigaOS support fixed.
912 * Revision 1.5 2001/05/22 18:46:04 oes
914 * - Enabled filtering banners by size rather than URL
915 * by adding patterns that replace all standard banner
916 * sizes with the "Junkbuster" gif to the re_filterfile
918 * - Enabled filtering WebBugs by providing a pattern
919 * which kills all 1x1 images
921 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
922 * which is selected by the (nonstandard and therefore
923 * capital) letter 'U' in the option string.
924 * It causes the quantifiers to be ungreedy by default.
925 * Appending a ? turns back to greedy (!).
927 * - Added a new interceptor ijb-send-banner, which
928 * sends back the "Junkbuster" gif. Without imagelist or
929 * MSIE detection support, or if tinygif = 1, or the
930 * URL isn't recognized as an imageurl, a lame HTML
931 * explanation is sent instead.
933 * - Added new feature, which permits blocking remote
934 * script redirects and firing back a local redirect
936 * The feature is conditionally compiled, i.e. it
937 * can be disabled with --disable-fast-redirects,
938 * plus it must be activated by a "fast-redirects"
939 * line in the config file, has its own log level
940 * and of course wants to be displayed by show-proxy-args
941 * Note: Boy, all the #ifdefs in 1001 locations and
942 * all the fumbling with configure.in and acconfig.h
943 * were *way* more work than the feature itself :-(
945 * - Because a generic redirect template was needed for
946 * this, tinygif = 3 now uses the same.
948 * - Moved GIFs, and other static HTTP response templates
953 * - Removed some >400 CRs again (Jon, you really worked
956 * Revision 1.4 2001/05/21 19:34:01 jongfoster
957 * Made failure to bind() a fatal error.
959 * Revision 1.3 2001/05/20 01:21:20 jongfoster
960 * Version 2.9.4 checkin.
961 * - Merged popupfile and cookiefile, and added control over PCRS
962 * filtering, in new "permissionsfile".
963 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
964 * file error you now get a message box (in the Win32 GUI) rather
965 * than the program exiting with no explanation.
966 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
968 * - Removed tabs from "config"
969 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
970 * - Bumped up version number.
972 * Revision 1.2 2001/05/17 22:34:44 oes
973 * - Added hint on GIF char array generation to jcc.c
974 * - Cleaned CRLF's from the sources and related files
975 * - Repaired logging for REF and FRC
977 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
978 * Initial import of version 2.9.3 source tree
981 *********************************************************************/
987 #include <sys/types.h>
996 # ifndef FEATURE_PTHREAD
1000 # include <windows.h>
1001 # include <process.h>
1002 # endif /* ndef FEATURE_PTHREAD */
1005 # ifndef _WIN_CONSOLE
1006 # include "w32log.h"
1007 # endif /* ndef _WIN_CONSOLE */
1008 # include "w32svrapi.h"
1010 #else /* ifndef _WIN32 */
1012 # if !defined (__OS2__)
1013 # include <unistd.h>
1014 # include <sys/wait.h>
1015 # endif /* ndef __OS2__ */
1016 # include <sys/time.h>
1017 # include <sys/stat.h>
1018 # include <sys/ioctl.h>
1021 #include <sys/termios.h>
1029 # include <signal.h>
1032 # include <socket.h> /* BeOS has select() for sockets only. */
1033 # include <OS.h> /* declarations for threads and stuff. */
1036 # if defined(__EMX__) || defined(__OS2__)
1037 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1042 #define bzero(B,N) memset(B,0x00,n)
1046 # include <select.h>
1051 #include "project.h"
1054 #include "filters.h"
1055 #include "loaders.h"
1056 #include "parsers.h"
1057 #include "miscutil.h"
1059 #include "jbsockets.h"
1060 #include "gateway.h"
1061 #include "actions.h"
1063 #include "loadcfg.h"
1064 #include "urlmatch.h"
1066 const char jcc_h_rcs[] = JCC_H_VERSION;
1067 const char project_h_rcs[] = PROJECT_H_VERSION;
1070 struct client_state clients[1];
1071 struct file_list files[1];
1073 #ifdef FEATURE_STATISTICS
1074 int urls_read = 0; /* total nr of urls read inc rejected */
1075 int urls_rejected = 0; /* total nr of urls rejected */
1076 #endif /* def FEATURE_STATISTICS */
1078 #ifdef FEATURE_GRACEFUL_TERMINATION
1079 int g_terminate = 0;
1082 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1083 static void sig_handler(int the_signal);
1085 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1086 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1087 static jb_err get_server_headers(struct client_state *csp);
1088 static const char *crunch_reason(const struct http_response *rsp);
1089 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
1091 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1093 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1094 static jb_err change_request_destination(struct client_state *csp);
1095 static void chat(struct client_state *csp);
1096 static void serve(struct client_state *csp);
1097 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1098 static void usage(const char *myname);
1100 static void initialize_mutexes(void);
1101 static jb_socket bind_port_helper(struct configuration_spec *config);
1102 static void listen_loop(void);
1105 void serve(struct client_state *csp);
1106 #else /* ifndef AMIGA */
1107 static void serve(struct client_state *csp);
1108 #endif /* def AMIGA */
1111 static int32 server_thread(void *data);
1112 #endif /* def __BEOS__ */
1115 #define sleep(N) Sleep(((N) * 1000))
1119 #define sleep(N) DosSleep(((N) * 100))
1122 #ifdef FEATURE_PTHREAD
1123 pthread_mutex_t log_mutex;
1124 pthread_mutex_t log_init_mutex;
1126 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1127 pthread_mutex_t resolver_mutex;
1128 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1130 #ifndef HAVE_GMTIME_R
1131 pthread_mutex_t gmtime_mutex;
1132 #endif /* ndef HAVE_GMTIME_R */
1134 #ifndef HAVE_LOCALTIME_R
1135 pthread_mutex_t localtime_mutex;
1136 #endif /* ndef HAVE_GMTIME_R */
1139 pthread_mutex_t rand_mutex;
1140 #endif /* ndef HAVE_RANDOM */
1142 #endif /* FEATURE_PTHREAD */
1144 #if defined(unix) || defined(__EMX__)
1145 const char *basedir = NULL;
1146 const char *pidfile = NULL;
1147 int received_hup_signal = 0;
1148 #endif /* defined unix */
1150 /* HTTP snipplets. */
1151 static const char CSUCCEED[] =
1152 "HTTP/1.0 200 Connection established\r\n"
1153 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1155 static const char CHEADER[] =
1156 "HTTP/1.0 400 Invalid header received from client\r\n"
1157 "Proxy-Agent: Privoxy " VERSION "\r\n"
1158 "Content-Type: text/plain\r\n"
1159 "Connection: close\r\n\r\n"
1160 "Invalid header received from client.\r\n";
1162 static const char FTP_RESPONSE[] =
1163 "HTTP/1.0 400 Invalid request received from client\r\n"
1164 "Content-Type: text/plain\r\n"
1165 "Connection: close\r\n\r\n"
1166 "Invalid request. Privoxy doesn't support FTP.\r\n";
1168 static const char GOPHER_RESPONSE[] =
1169 "HTTP/1.0 400 Invalid request received from client\r\n"
1170 "Content-Type: text/plain\r\n"
1171 "Connection: close\r\n\r\n"
1172 "Invalid request. Privoxy doesn't support gopher.\r\n";
1174 /* XXX: should be a template */
1175 static const char MISSING_DESTINATION_RESPONSE[] =
1176 "HTTP/1.0 400 Bad request received from client\r\n"
1177 "Proxy-Agent: Privoxy " VERSION "\r\n"
1178 "Content-Type: text/plain\r\n"
1179 "Connection: close\r\n\r\n"
1180 "Bad request. Privoxy was unable to extract the destination.\r\n";
1182 /* XXX: should be a template */
1183 static const char NO_SERVER_DATA_RESPONSE[] =
1184 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1185 "Proxy-Agent: Privoxy " VERSION "\r\n"
1186 "Content-Type: text/plain\r\n"
1187 "Connection: close\r\n\r\n"
1188 "Empty server or forwarder response.\r\n"
1189 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1191 /* XXX: should be a template */
1192 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1193 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1194 "Proxy-Agent: Privoxy " VERSION "\r\n"
1195 "Content-Type: text/plain\r\n"
1196 "Connection: close\r\n\r\n"
1197 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1200 /* XXX: should be a template */
1201 static const char NULL_BYTE_RESPONSE[] =
1202 "HTTP/1.0 400 Bad request received from client\r\n"
1203 "Proxy-Agent: Privoxy " VERSION "\r\n"
1204 "Content-Type: text/plain\r\n"
1205 "Connection: close\r\n\r\n"
1206 "Bad request. Null byte(s) before end of request.\r\n";
1209 /* XXX: should be a template */
1210 static const char MESSED_UP_REQUEST_RESPONSE[] =
1211 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1212 "Proxy-Agent: Privoxy " VERSION "\r\n"
1213 "Content-Type: text/plain\r\n"
1214 "Connection: close\r\n\r\n"
1215 "Bad request. Messed up with header filters.\r\n";
1217 /* A function to crunch a response */
1218 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1220 /* Crunch function flags */
1221 #define CF_NO_FLAGS 0
1222 /* Cruncher applies to forced requests as well */
1223 #define CF_IGNORE_FORCE 1
1224 /* Crunched requests are counted for the block statistics */
1225 #define CF_COUNT_AS_REJECT 2
1227 /* A crunch function and its flags */
1230 const crunch_func_ptr cruncher;
1234 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1236 /* Complete list of cruncher functions */
1237 static const struct cruncher crunchers_all[] = {
1238 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1239 { block_url, CF_COUNT_AS_REJECT },
1240 #ifdef FEATURE_TRUST
1241 { trust_url, CF_COUNT_AS_REJECT },
1242 #endif /* def FEATURE_TRUST */
1243 { redirect_url, CF_NO_FLAGS },
1244 { dispatch_cgi, CF_IGNORE_FORCE},
1248 /* Light version, used after tags are applied */
1249 static const struct cruncher crunchers_light[] = {
1250 { block_url, CF_COUNT_AS_REJECT },
1251 { redirect_url, CF_NO_FLAGS },
1256 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1257 /*********************************************************************
1259 * Function : sig_handler
1261 * Description : Signal handler for different signals.
1262 * Exit gracefully on TERM and INT
1263 * or set a flag that will cause the errlog
1264 * to be reopened by the main thread on HUP.
1267 * 1 : the_signal = the signal cause this function to call
1271 *********************************************************************/
1272 static void sig_handler(int the_signal)
1278 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1289 received_hup_signal = 1;
1294 * We shouldn't be here, unless we catch signals
1295 * in main() that we can't handle here!
1297 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1305 /*********************************************************************
1307 * Function : client_protocol_is_unsupported
1309 * Description : Checks if the client used a known unsupported
1310 * protocol and deals with it by sending an error
1314 * 1 : csp = Current client state (buffers, headers, etc...)
1315 * 2 : req = the first request line send by the client
1317 * Returns : TRUE if an error response has been generated, or
1318 * FALSE if the request doesn't look invalid.
1320 *********************************************************************/
1321 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1323 char buf[BUFFER_SIZE];
1326 * If it's a FTP or gopher request, we don't support it.
1328 * These checks are better than nothing, but they might
1329 * not work in all configurations and some clients might
1330 * have problems digesting the answer.
1332 * They should, however, never cause more problems than
1333 * Privoxy's old behaviour (returning the misleading HTML
1336 * "Could not resolve http://(ftp|gopher)://example.org").
1338 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1340 if (!strncmpic(req, "GET ftp://", 10))
1342 strlcpy(buf, FTP_RESPONSE, sizeof(buf));
1343 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1344 csp->ip_addr_str, req);
1348 strlcpy(buf, GOPHER_RESPONSE, sizeof(buf));
1349 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1350 csp->ip_addr_str, req);
1352 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1354 write_socket(csp->cfd, buf, strlen(buf));
1363 /*********************************************************************
1365 * Function : get_request_destination_elsewhere
1367 * Description : If the client's request was redirected into
1368 * Privoxy without the client's knowledge,
1369 * the request line lacks the destination host.
1371 * This function tries to get it elsewhere,
1372 * provided accept-intercepted-requests is enabled.
1374 * "Elsewhere" currently only means "Host: header",
1375 * but in the future we may ask the redirecting
1376 * packet filter to look the destination up.
1378 * If the destination stays unknown, an error
1379 * response is send to the client and headers
1380 * are freed so that chat() can return directly.
1383 * 1 : csp = Current client state (buffers, headers, etc...)
1384 * 2 : headers = a header list
1386 * Returns : JB_ERR_OK if the destination is now known, or
1387 * JB_ERR_PARSE if it isn't.
1389 *********************************************************************/
1390 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1394 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1396 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1397 " Privoxy isn't configured to accept intercepted requests.",
1398 csp->ip_addr_str, csp->http->cmd);
1399 /* XXX: Use correct size */
1400 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1401 csp->ip_addr_str, csp->http->cmd);
1403 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1404 destroy_list(headers);
1406 return JB_ERR_PARSE;
1408 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1410 /* Split the domain we just got for pattern matching */
1411 init_domain_components(csp->http);
1417 /* We can't work without destination. Go spread the news.*/
1419 req = list_to_text(headers);
1421 /* XXX: Use correct size */
1422 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1423 csp->ip_addr_str, csp->http->cmd);
1424 log_error(LOG_LEVEL_ERROR,
1425 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1426 csp->ip_addr_str, csp->http->cmd, req);
1429 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1430 destroy_list(headers);
1432 return JB_ERR_PARSE;
1435 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1436 * to get the destination IP address, use it as host directly
1437 * or do a reverse DNS lookup first.
1442 /*********************************************************************
1444 * Function : get_server_headers
1446 * Description : Parses server headers in iob and fills them
1447 * into csp->headers so that they can later be
1451 * 1 : csp = Current client state (buffers, headers, etc...)
1453 * Returns : JB_ERR_OK if everything went fine, or
1454 * JB_ERR_PARSE if the headers were incomplete.
1456 *********************************************************************/
1457 static jb_err get_server_headers(struct client_state *csp)
1459 int continue_hack_in_da_house = 0;
1462 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1467 * continue hack in da house. Ignore the ending of
1468 * this head and continue enlisting header lines.
1469 * The reason is described below.
1471 enlist(csp->headers, "");
1472 continue_hack_in_da_house = 0;
1475 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1478 * It's a bodyless continue response, don't
1479 * stop header parsing after reaching it's end.
1481 * As a result Privoxy will concatenate the
1482 * next response's head and parse and deliver
1483 * the headers as if they belonged to one request.
1485 * The client will separate them because of the
1486 * empty line between them.
1488 * XXX: What we're doing here is clearly against
1489 * the intended purpose of the continue header,
1490 * and under some conditions (HTTP/1.0 client request)
1491 * it's a standard violation.
1493 * Anyway, "sort of against the spec" is preferable
1494 * to "always getting confused by Continue responses"
1495 * (Privoxy's behaviour before this hack was added)
1497 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1498 continue_hack_in_da_house = 1;
1500 else if (*header == '\0')
1503 * If the header is empty, but the Continue hack
1504 * isn't active, we can assume that we reached the
1505 * end of the buffer before we hit the end of the
1508 * Inform the caller an let it decide how to handle it.
1510 return JB_ERR_PARSE;
1514 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1517 * XXX: Should we quit the request and return a
1518 * out of memory error page instead?
1520 log_error(LOG_LEVEL_ERROR,
1521 "Out of memory while enlisting server headers. %s lost.",
1531 /*********************************************************************
1533 * Function : crunch_reason
1535 * Description : Translates the crunch reason code into a string.
1538 * 1 : rsp = a http_response
1540 * Returns : A string with the crunch reason or an error description.
1542 *********************************************************************/
1543 static const char *crunch_reason(const struct http_response *rsp)
1545 char * reason = NULL;
1547 assert(rsp != NULL);
1550 return "Internal error while searching for crunch reason";
1553 switch (rsp->reason)
1555 case RSP_REASON_UNSUPPORTED:
1556 reason = "Unsupported HTTP feature";
1558 case RSP_REASON_BLOCKED:
1561 case RSP_REASON_UNTRUSTED:
1562 reason = "Untrusted";
1564 case RSP_REASON_REDIRECTED:
1565 reason = "Redirected";
1567 case RSP_REASON_CGI_CALL:
1568 reason = "CGI Call";
1570 case RSP_REASON_NO_SUCH_DOMAIN:
1571 reason = "DNS failure";
1573 case RSP_REASON_FORWARDING_FAILED:
1574 reason = "Forwarding failed";
1576 case RSP_REASON_CONNECT_FAILED:
1577 reason = "Connection failure";
1579 case RSP_REASON_OUT_OF_MEMORY:
1580 reason = "Out of memory (may mask other reasons)";
1583 reason = "No reason recorded";
1591 /*********************************************************************
1593 * Function : send_crunch_response
1595 * Description : Delivers already prepared response for
1596 * intercepted requests, logs the interception
1597 * and frees the response.
1600 * 1 : csp = Current client state (buffers, headers, etc...)
1601 * 1 : rsp = Fully prepared response. Will be freed on exit.
1603 * Returns : Nothing.
1605 *********************************************************************/
1606 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1608 const struct http_request *http = csp->http;
1609 char status_code[4];
1611 assert(rsp != NULL);
1612 assert(rsp->head != NULL);
1617 * Not supposed to happen. If it does
1618 * anyway, treat it as an unknown error.
1620 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1621 /* return code doesn't matter */
1626 /* If rsp is still NULL, we have serious internal problems. */
1627 log_error(LOG_LEVEL_FATAL,
1628 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1632 * Extract the status code from the actual head
1633 * that was send to the client. It is the only
1634 * way to get it right for all requests, including
1635 * the fixed ones for out-of-memory problems.
1637 * A head starts like this: 'HTTP/1.1 200...'
1641 status_code[0] = rsp->head[9];
1642 status_code[1] = rsp->head[10];
1643 status_code[2] = rsp->head[11];
1644 status_code[3] = '\0';
1646 /* Write the answer to the client */
1647 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1648 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1650 /* There is nothing we can do about it. */
1651 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1654 /* Log that the request was crunched and why. */
1655 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1656 http->hostport, http->path, crunch_reason(rsp));
1657 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1658 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1660 /* Clean up and return */
1661 if (cgi_error_memory() != rsp)
1663 free_http_response(rsp);
1670 /*********************************************************************
1672 * Function : request_contains_null_bytes
1674 * Description : Checks for NULL bytes in the request and sends
1675 * an error message to the client if any were found.
1677 * XXX: currently not used, see comment in chat().
1680 * 1 : csp = Current client state (buffers, headers, etc...)
1681 * 2 : buf = Data from the client's request to check.
1682 * 3 : len = The data length.
1684 * Returns : TRUE if the request contained one or more NULL bytes, or
1687 *********************************************************************/
1688 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1690 size_t c_len; /* Request lenght when treated as C string */
1692 c_len = strlen(buf);
1697 * Null byte(s) found. Log the request,
1698 * return an error response and hang up.
1700 size_t tmp_len = c_len;
1705 * Replace NULL byte(s) with '°' characters
1706 * so the request can be logged as string.
1707 * XXX: Is there a better replacement character?
1710 tmp_len += strlen(buf+tmp_len);
1711 } while (tmp_len < len);
1713 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1714 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1715 log_error(LOG_LEVEL_HEADER,
1716 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1718 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1720 /* XXX: Log correct size */
1721 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1731 /*********************************************************************
1733 * Function : crunch_response_triggered
1735 * Description : Checks if the request has to be crunched,
1736 * and delivers the crunch response if necessary.
1739 * 1 : csp = Current client state (buffers, headers, etc...)
1740 * 2 : crunchers = list of cruncher functions to run
1742 * Returns : TRUE if the request was answered with a crunch response
1745 *********************************************************************/
1746 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1748 struct http_response *rsp = NULL;
1749 const struct cruncher *c;
1752 * If CGI request crunching is disabled,
1753 * check the CGI dispatcher out of order to
1754 * prevent unintentional blocks or redirects.
1756 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1757 && (NULL != (rsp = dispatch_cgi(csp))))
1759 /* Deliver, log and free the interception response. */
1760 send_crunch_response(csp, rsp);
1764 for (c = crunchers; c->cruncher != NULL; c++)
1767 * Check the cruncher if either Privoxy is toggled
1768 * on and the request isn't forced, or if the cruncher
1769 * applies to forced requests as well.
1771 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1772 !(csp->flags & CSP_FLAG_FORCED)) ||
1773 (c->flags & CF_IGNORE_FORCE))
1775 rsp = c->cruncher(csp);
1778 /* Deliver, log and free the interception response. */
1779 send_crunch_response(csp, rsp);
1780 #ifdef FEATURE_STATISTICS
1781 if (c->flags & CF_COUNT_AS_REJECT)
1783 csp->flags |= CSP_FLAG_REJECTED;
1785 #endif /* def FEATURE_STATISTICS */
1796 /*********************************************************************
1798 * Function : build_request_line
1800 * Description : Builds the HTTP request line.
1802 * If a HTTP forwarder is used it expects the whole URL,
1803 * web servers only get the path.
1806 * 1 : csp = Current client state (buffers, headers, etc...)
1807 * 2 : fwd = The forwarding spec used for the request
1808 * XXX: Should use http->fwd instead.
1809 * 3 : request_line = The old request line which will be replaced.
1811 * Returns : Nothing. Terminates in case of memory problems.
1813 *********************************************************************/
1814 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1816 struct http_request *http = csp->http;
1818 assert(http->ssl == 0);
1821 * Downgrade http version from 1.1 to 1.0
1822 * if +downgrade action applies.
1824 if ( (csp->action->flags & ACTION_DOWNGRADE)
1825 && (!strcmpic(http->ver, "HTTP/1.1")))
1828 http->ver = strdup("HTTP/1.0");
1830 if (http->ver == NULL)
1832 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1837 * Rebuild the request line.
1839 freez(*request_line);
1840 *request_line = strdup(http->gpc);
1841 string_append(request_line, " ");
1843 if (fwd->forward_host)
1845 string_append(request_line, http->url);
1849 string_append(request_line, http->path);
1851 string_append(request_line, " ");
1852 string_append(request_line, http->ver);
1854 if (*request_line == NULL)
1856 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1858 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1862 /*********************************************************************
1864 * Function : change_request_destination
1866 * Description : Parse a (rewritten) request line and regenerate
1867 * the http request data.
1870 * 1 : csp = Current client state (buffers, headers, etc...)
1872 * Returns : Forwards the parse_http_request() return code.
1873 * Terminates in case of memory problems.
1875 *********************************************************************/
1876 static jb_err change_request_destination(struct client_state *csp)
1878 struct http_request *http = csp->http;
1881 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1882 free_http_request(http);
1883 err = parse_http_request(csp->headers->first->str, http, csp);
1884 if (JB_ERR_OK != err)
1886 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1887 jb_err_to_string(err));
1889 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1890 if (http->ocmd == NULL)
1892 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1899 /*********************************************************************
1903 * Description : Once a connection to the client has been accepted,
1904 * this function is called (via serve()) to handle the
1905 * main business of the communication. When this
1906 * function returns, the caller must close the client
1909 * FIXME: chat is nearly thousand lines long.
1913 * 1 : csp = Current client state (buffers, headers, etc...)
1915 * Returns : Nothing.
1917 *********************************************************************/
1918 static void chat(struct client_state *csp)
1920 char buf[BUFFER_SIZE];
1928 int ms_iis5_hack = 0;
1929 size_t byte_count = 0;
1930 int forwarded_connect_retries = 0;
1931 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1932 const struct forward_spec * fwd;
1933 struct http_request *http;
1934 int len; /* for buffer sizes (and negative error codes) */
1937 /* Function that does the content filtering for the current request */
1938 filter_function_ptr content_filter = NULL;
1940 /* Skeleton for HTTP response, if we should intercept the request */
1941 struct http_response *rsp;
1943 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1944 struct list header_list;
1945 struct list *headers = &header_list;
1949 memset(buf, 0, sizeof(buf));
1952 * Read the client's request. Note that since we're not using select() we
1953 * could get blocked here if a client connected, then didn't say anything!
1958 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1960 if (len <= 0) break; /* error! */
1963 * If there is no memory left for buffering the
1964 * request, there is nothing we can do but hang up
1966 if (add_to_iob(csp, buf, len))
1971 req = get_header(csp);
1973 } while ((NULL != req) && ('\0' == *req));
1975 if ((NULL != req) && ('\0' != *req))
1977 /* Request received. Validate and parse it. */
1981 * XXX: Temporary disabled to prevent problems
1982 * with POST requests whose bodies are allowed to
1983 * contain NULL bytes. BR#1730105.
1985 * The main purpose of this check is to properly
1986 * log stuff like BitTorrent traffic and other junk
1987 * that hits public proxies. It's not required for
1988 * Privoxy to functions as those requests are discarded
1991 * It probably should be rewritten to only check
1992 * the head of the request. Another option would
1993 * be to let all POST requests pass, although that
1994 * may not be good enough.
1996 if (request_contains_null_bytes(csp, buf, len))
1998 /* NULL bytes found and dealt with, just hang up. */
2003 /* Does the request line look invalid? */
2004 if (client_protocol_is_unsupported(csp, req))
2007 * Yes. The request has already been
2008 * answered with a error response, the buffers
2009 * were freed and we're done with chatting.
2014 #ifdef FEATURE_FORCE_LOAD
2016 * If this request contains the FORCE_PREFIX and blocks
2017 * aren't enforced, get rid of it and set the force flag.
2019 if (strstr(req, FORCE_PREFIX))
2021 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2023 log_error(LOG_LEVEL_FORCE,
2024 "Ignored force prefix in request: \"%s\".", req);
2028 strclean(req, FORCE_PREFIX);
2029 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2030 csp->flags |= CSP_FLAG_FORCED;
2034 #endif /* def FEATURE_FORCE_LOAD */
2035 err = parse_http_request(req, http, csp);
2036 if (JB_ERR_OK != err)
2038 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2044 if (http->cmd == NULL)
2046 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2047 /* XXX: Use correct size */
2048 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2049 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2051 free_http_request(http);
2055 /* grab the rest of the client's headers */
2059 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
2061 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2064 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2065 destroy_list(headers);
2070 * If there is no memory left for buffering the
2071 * request, there is nothing we can do but hang up
2073 if (add_to_iob(csp, buf, len))
2075 destroy_list(headers);
2081 if (p == NULL) break;
2088 if (http->host == NULL)
2091 * If we still don't know the request destination,
2092 * the request is invalid or the client uses
2093 * Privoxy without its knowledge.
2095 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2098 * Our attempts to get the request destination
2099 * elsewhere failed or Privoxy is configured
2100 * to only accept proxy requests.
2102 * An error response has already been send
2103 * and we're done here.
2110 * Determine the actions for this URL
2112 #ifdef FEATURE_TOGGLE
2113 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2115 /* Most compatible set of actions (i.e. none) */
2116 init_current_action(csp->action);
2119 #endif /* ndef FEATURE_TOGGLE */
2121 get_url_actions(csp, http);
2125 * Save a copy of the original request for logging
2127 http->ocmd = strdup(http->cmd);
2129 if (http->ocmd == NULL)
2131 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2134 enlist(csp->headers, http->cmd);
2136 /* Append the previously read headers */
2137 list_append_list_unique(csp->headers, headers);
2138 destroy_list(headers);
2140 err = sed(client_patterns, add_client_headers, csp);
2141 if (JB_ERR_OK != err)
2143 assert(err == JB_ERR_PARSE);
2144 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2146 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2148 /* Check request line for rewrites. */
2149 if ((NULL == csp->headers->first->str)
2150 || (strcmp(http->cmd, csp->headers->first->str) &&
2151 (JB_ERR_OK != change_request_destination(csp))))
2154 * A header filter broke the request line - bail out.
2156 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2157 /* XXX: Use correct size */
2158 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2159 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2161 free_http_request(http);
2165 /* decide how to route the HTTP request */
2166 fwd = forward_url(csp, http);
2169 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2170 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2174 * build the http request to send to the server
2175 * we have to do one of the following:
2177 * create = use the original HTTP request to create a new
2178 * HTTP request that has either the path component
2179 * without the http://domainspec (w/path) or the
2180 * full orininal URL (w/url)
2181 * Note that the path and/or the HTTP version may
2182 * have been altered by now.
2184 * connect = Open a socket to the host:port of the server
2185 * and short-circuit server and client socket.
2187 * pass = Pass the request unchanged if forwarding a CONNECT
2188 * request to a parent proxy. Note that we'll be sending
2189 * the CFAIL message ourselves if connecting to the parent
2190 * fails, but we won't send a CSUCCEED message if it works,
2191 * since that would result in a double message (ours and the
2192 * parent's). After sending the request to the parent, we simply
2195 * here's the matrix:
2198 * +--------+--------+
2200 * 0 | create | connect|
2202 * Forwarding +--------+--------+
2204 * 1 | create | pass |
2206 * +--------+--------+
2210 if (http->ssl && connect_port_is_forbidden(csp))
2212 const char *acceptable_connect_ports =
2213 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2214 assert(NULL != acceptable_connect_ports);
2215 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2216 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2217 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2218 csp->action->flags |= ACTION_BLOCK;
2224 freez(csp->headers->first->str);
2225 build_request_line(csp, fwd, &csp->headers->first->str);
2228 hdr = list_to_text(csp->headers);
2231 /* FIXME Should handle error properly */
2232 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2236 * We have a request. Check if one of the crunchers wants it.
2238 if (crunch_response_triggered(csp, crunchers_all))
2241 * Yes. The client got the crunch response
2242 * and we are done here after cleaning up.
2245 list_remove_all(csp->headers);
2251 * The headers can't be removed earlier because
2252 * they were still needed for the referrer check
2253 * in case of CGI crunches.
2255 * XXX: Would it be worth to move the referrer check
2256 * into client_referrer() and set a flag if it's trusted?
2258 list_remove_all(csp->headers);
2260 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2262 if (fwd->forward_host)
2264 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2265 fwd->forward_host, fwd->forward_port, http->hostport);
2269 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2272 /* here we connect to the server, gateway, or the forwarder */
2274 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2275 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2277 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2278 forwarded_connect_retries, http->hostport);
2281 if (csp->sfd == JB_INVALID_SOCKET)
2283 if (fwd->type != SOCKS_NONE)
2286 rsp = error_response(csp, "forwarding-failed", errno);
2288 else if (errno == EINVAL)
2290 rsp = error_response(csp, "no-such-domain", errno);
2294 rsp = error_response(csp, "connect-failed", errno);
2295 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2300 /* Write the answer to the client */
2303 send_crunch_response(csp, rsp);
2310 if (fwd->forward_host || (http->ssl == 0))
2312 /* write the client's (modified) header to the server
2313 * (along with anything else that may be in the buffer)
2316 if (write_socket(csp->sfd, hdr, strlen(hdr))
2317 || (flush_socket(csp->sfd, csp->iob) < 0))
2319 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2322 rsp = error_response(csp, "connect-failed", errno);
2326 send_crunch_response(csp, rsp);
2336 * We're running an SSL tunnel and we're not forwarding,
2337 * so just send the "connect succeeded" message to the
2338 * client, flush the rest, and get out of the way.
2340 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2348 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2350 /* we're finished with the client's header */
2353 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2355 /* pass data between the client and server
2356 * until one or the other shuts down the connection.
2365 * FD_ZERO here seems to point to an errant macro which crashes.
2366 * So do this by hand for now...
2368 memset(&rfds,0x00,sizeof(fd_set));
2372 FD_SET(csp->cfd, &rfds);
2373 FD_SET(csp->sfd, &rfds);
2375 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2379 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2383 /* this is the body of the browser's request
2384 * just read it and write it.
2387 if (FD_ISSET(csp->cfd, &rfds))
2389 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2393 break; /* "game over, man" */
2396 if (write_socket(csp->sfd, buf, (size_t)len))
2398 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2405 * The server wants to talk. It could be the header or the body.
2406 * If `hdr' is null, then it's the header otherwise it's the body.
2407 * FIXME: Does `hdr' really mean `host'? No.
2411 if (FD_ISSET(csp->sfd, &rfds))
2414 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2418 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2420 if (http->ssl && (fwd->forward_host == NULL))
2423 * Just hang up. We already confirmed the client's CONNECT
2424 * request with status code 200 and unencrypted content is
2425 * no longer welcome.
2427 log_error(LOG_LEVEL_ERROR,
2428 "CONNECT already confirmed. Unable to tell the client about the problem.");
2431 else if (byte_count)
2434 * Just hang up. We already transmitted the original headers
2435 * and parts of the original content and therefore missed the
2436 * chance to send an error message (without risking data corruption).
2438 * XXX: we could retry with a fancy range request here.
2440 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2441 "Unable to tell the client about the problem.");
2445 rsp = error_response(csp, "connect-failed", errno);
2449 send_crunch_response(csp, rsp);
2455 /* Add a trailing zero. This lets filter_popups
2456 * use string operations.
2460 /* Normally, this would indicate that we've read
2461 * as much as the server has sent us and we can
2462 * close the client connection. However, Microsoft
2463 * in its wisdom has released IIS/5 with a bug that
2464 * prevents it from sending the trailing \r\n in
2465 * a 302 redirect header (and possibly other headers).
2466 * To work around this if we've haven't parsed
2467 * a full header we'll append a trailing \r\n
2468 * and see if this now generates a valid one.
2470 * This hack shouldn't have any impacts. If we've
2471 * already transmitted the header or if this is a
2472 * SSL connection, then we won't bother with this
2473 * hack. So we only work on partially received
2474 * headers. If we append a \r\n and this still
2475 * doesn't generate a valid header, then we won't
2476 * transmit anything to the client.
2481 if (server_body || http->ssl)
2484 * If we have been buffering up the document,
2485 * now is the time to apply content modification
2486 * and send the result to the client.
2490 p = execute_content_filter(csp, content_filter);
2492 * If the content filter fails, use the original
2493 * buffer and length.
2494 * (see p != NULL ? p : csp->iob->cur below)
2498 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2501 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2503 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2506 hdr = list_to_text(csp->headers);
2509 /* FIXME Should handle error properly */
2510 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2514 * Shouldn't happen because this was the second sed run
2515 * and tags are only created for the first one.
2517 assert(!crunch_response_triggered(csp, crunchers_all));
2519 if (write_socket(csp->cfd, hdr, strlen(hdr))
2520 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2522 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2532 break; /* "game over, man" */
2536 * This is NOT the body, so
2537 * Let's pretend the server just sent us a blank line.
2539 log_error(LOG_LEVEL_INFO,
2540 "Malformerd HTTP headers detected and MS IIS5 hack enabled. "
2541 "Expect an invalid response or even no response at all.");
2542 snprintf(buf, sizeof(buf), "\r\n");
2543 len = (int)strlen(buf);
2546 * Now, let the normal header parsing algorithm below do its
2547 * job. If it fails, we'll exit instead of continuing.
2554 * If this is an SSL connection or we're in the body
2555 * of the server document, just write it to the client,
2556 * unless we need to buffer the body for later content-filtering
2559 if (server_body || http->ssl)
2564 * If there is no memory left for buffering the content, or the buffer limit
2565 * has been reached, switch to non-filtering mode, i.e. make & write the
2566 * header, flush the iob and buf, and get out of the way.
2568 if (add_to_iob(csp, buf, len))
2573 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2575 hdr = list_to_text(csp->headers);
2579 * Memory is too tight to even generate the header.
2580 * Send our static "Out-of-memory" page.
2582 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2583 rsp = cgi_error_memory();
2584 send_crunch_response(csp, rsp);
2588 hdrlen = strlen(hdr);
2590 if (write_socket(csp->cfd, hdr, hdrlen)
2591 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2592 || (write_socket(csp->cfd, buf, (size_t)len)))
2594 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2601 * Reset the byte_count to the amount of bytes
2602 * we just flushed. len will be added a few lines below,
2603 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2605 byte_count = (size_t)flushed;
2607 content_filter = NULL;
2613 if (write_socket(csp->cfd, buf, (size_t)len))
2615 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2619 byte_count += (size_t)len;
2624 /* we're still looking for the end of the
2625 * server's header ... (does that make header
2626 * parsing an "out of body experience" ?
2630 * buffer up the data we just read. If that fails,
2631 * there's little we can do but send our static
2632 * out-of-memory page.
2634 if (add_to_iob(csp, buf, len))
2636 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2637 rsp = cgi_error_memory();
2638 send_crunch_response(csp, rsp);
2643 /* Convert iob into something sed() can digest */
2644 if (JB_ERR_PARSE == get_server_headers(csp))
2648 /* Well, we tried our MS IIS/5
2649 * hack and it didn't work.
2650 * The header is incomplete
2651 * and there isn't anything
2652 * we can do about it.
2658 /* Since we have to wait for
2659 * more from the server before
2660 * we can parse the headers
2661 * we just continue here.
2667 /* Did we actually get anything? */
2668 if (NULL == csp->headers->first)
2670 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2671 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2672 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2673 free_http_request(http);
2677 assert(csp->headers->first->str);
2679 if (strncmpic(csp->headers->first->str, "HTTP", 4))
2682 * It doesn't look like a HTTP response:
2683 * tell the client and log the problem.
2685 if (strlen(csp->headers->first->str) > 30)
2687 csp->headers->first->str[30] = '\0';
2689 log_error(LOG_LEVEL_ERROR,
2690 "Invalid server or forwarder response. Starts with: %s",
2691 csp->headers->first->str);
2692 log_error(LOG_LEVEL_CLF,
2693 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2694 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2695 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2696 free_http_request(http);
2700 /* we have now received the entire header.
2701 * filter it and send the result to the client
2703 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2705 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2707 hdr = list_to_text(csp->headers);
2710 /* FIXME Should handle error properly */
2711 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2714 if (crunch_response_triggered(csp, crunchers_light))
2717 * One of the tags created by a server-header
2718 * tagger triggered a crunch. We already
2719 * delivered the crunch response to the client
2720 * and are done here after cleaning up.
2725 /* Buffer and pcrs filter this if appropriate. */
2727 if (!http->ssl) /* We talk plaintext */
2729 content_filter = get_filter_function(csp);
2732 * Only write if we're not buffering for content modification
2734 if (!content_filter)
2736 /* write the server's (modified) header to
2737 * the client (along with anything else that
2738 * may be in the buffer)
2741 if (write_socket(csp->cfd, hdr, strlen(hdr))
2742 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2744 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2746 /* the write failed, so don't bother
2747 * mentioning it to the client...
2748 * it probably can't hear us anyway.
2754 byte_count += (size_t)len;
2757 /* we're finished with the server's header */
2762 /* If this was a MS IIS/5 hack then it means
2763 * the server has already closed the
2764 * connection. Nothing more to read. Time
2775 return; /* huh? we should never get here */
2778 if (csp->content_length == 0)
2781 * If Privoxy didn't recalculate the
2782 * Content-Lenght, byte_count is still
2785 csp->content_length = byte_count;
2788 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2789 csp->ip_addr_str, http->ocmd, csp->content_length);
2793 /*********************************************************************
2797 * Description : This is little more than chat. We only "serve" to
2798 * to close any socket that chat may have opened.
2801 * 1 : csp = Current client state (buffers, headers, etc...)
2805 *********************************************************************/
2807 void serve(struct client_state *csp)
2808 #else /* ifndef AMIGA */
2809 static void serve(struct client_state *csp)
2810 #endif /* def AMIGA */
2813 close_socket(csp->cfd);
2815 if (csp->sfd != JB_INVALID_SOCKET)
2817 close_socket(csp->sfd);
2820 csp->flags &= ~CSP_FLAG_ACTIVE;
2826 /*********************************************************************
2828 * Function : server_thread
2830 * Description : We only exist to call `serve' in a threaded environment.
2833 * 1 : data = Current client state (buffers, headers, etc...)
2835 * Returns : Always 0.
2837 *********************************************************************/
2838 static int32 server_thread(void *data)
2840 serve((struct client_state *) data);
2847 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2848 /*********************************************************************
2852 * Description : Print usage info & exit.
2854 * Parameters : Pointer to argv[0] for identifying ourselves
2858 *********************************************************************/
2859 static void usage(const char *myname)
2861 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2865 #endif /* defined(unix) */
2868 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2869 #endif /* defined(unix) */
2870 "[--version] [configfile]\n"
2871 "Aborting\n", myname);
2876 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2879 /*********************************************************************
2881 * Function : initialize_mutexes
2883 * Description : Prepares mutexes if mutex support is available.
2887 * Returns : Void, exits in case of errors.
2889 *********************************************************************/
2890 static void initialize_mutexes(void)
2894 #ifdef FEATURE_PTHREAD
2896 * Prepare global mutex semaphores
2898 err = pthread_mutex_init(&log_mutex, 0);
2900 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2903 * XXX: The assumptions below are a bit naive
2904 * and can cause locks that aren't necessary.
2906 * For example older FreeBSD versions (< 6.x?)
2907 * have no gethostbyname_r, but gethostbyname is
2910 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2911 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2912 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2914 * XXX: should we use a single mutex for
2915 * localtime() and gmtime() as well?
2917 #ifndef HAVE_GMTIME_R
2918 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2919 #endif /* ndef HAVE_GMTIME_R */
2921 #ifndef HAVE_LOCALTIME_R
2922 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2923 #endif /* ndef HAVE_GMTIME_R */
2926 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2927 #endif /* ndef HAVE_RANDOM */
2928 #endif /* FEATURE_PTHREAD */
2931 * TODO: mutex support for mingw32 would be swell.
2936 printf("Fatal error. Mutex initialization failed: %s.\n",
2945 /*********************************************************************
2949 * Description : Load the config file and start the listen loop.
2950 * This function is a lot more *sane* with the `load_config'
2951 * and `listen_loop' functions; although it stills does
2952 * a *little* too much for my taste.
2955 * 1 : argc = Number of parameters (including $0).
2956 * 2 : argv = Array of (char *)'s to the parameters.
2958 * Returns : 1 if : can't open config file, unrecognized directive,
2959 * stats requested in multi-thread mode, can't open the
2960 * log file, can't open the jar file, listen port is invalid,
2961 * any load fails, and can't bind port.
2963 * Else main never returns, the process must be signaled
2964 * to terminate execution. Or, on Windows, use the
2965 * "File", "Exit" menu option.
2967 *********************************************************************/
2969 int real_main(int argc, const char *argv[])
2971 int main(int argc, const char *argv[])
2975 unsigned int random_seed;
2977 struct passwd *pw = NULL;
2978 struct group *grp = NULL;
2981 char *pre_chroot_nslookup_to_load_resolver = NULL;
2988 #if !defined(_WIN32)
2996 * Parse the command line arguments
2998 * XXX: simply printing usage information in case of
2999 * invalid arguments isn't particular user friendly.
3001 while (++argc_pos < argc)
3004 /* Check to see if the service must be installed or uninstalled */
3005 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3007 const char *pName = argv[argc_pos] + 9;
3010 exit( (install_service(pName)) ? 0 : 1 );
3012 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3014 const char *pName = argv[argc_pos] + 11;
3017 exit((uninstall_service(pName)) ? 0 : 1);
3019 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3021 bRunAsService = TRUE;
3022 w32_set_service_cwd();
3023 atexit(w32_service_exit_notify);
3026 #endif /* defined(_WIN32) */
3029 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3031 if (strcmp(argv[argc_pos], "--help") == 0)
3036 else if(strcmp(argv[argc_pos], "--version") == 0)
3038 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3044 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3049 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3051 if (++argc_pos == argc) usage(argv[0]);
3052 pidfile = strdup(argv[argc_pos]);
3055 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3057 if (++argc_pos == argc) usage(argv[argc_pos]);
3059 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3062 if (NULL == (grp = getgrnam(p)))
3064 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3068 if (NULL == (pw = getpwnam(argv[argc_pos])))
3070 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3073 if (p != NULL) *--p = '\0';
3076 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3078 if (++argc_pos == argc) usage(argv[0]);
3079 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3082 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3086 #endif /* defined(unix) */
3088 else if (argc_pos + 1 != argc)
3091 * This is neither the last command line
3092 * option, nor was it recognized before,
3093 * therefore it must be invalid.
3099 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3101 configfile = argv[argc_pos];
3104 } /* -END- while (more arguments) */
3107 if ( *configfile != '/' )
3109 char cwd[BUFFER_SIZE];
3111 size_t abs_file_size;
3113 /* make config-filename absolute here */
3114 if (NULL == getcwd(cwd, sizeof(cwd)))
3116 perror("failed to get current working directory");
3121 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3122 basedir = strdup(cwd);
3124 if (NULL == basedir ||
3125 NULL == (abs_file = malloc(abs_file_size)))
3127 perror("malloc failed");
3130 strlcpy(abs_file, basedir, abs_file_size);
3131 strlcat(abs_file, "/", abs_file_size );
3132 strlcat(abs_file, configfile, abs_file_size);
3133 configfile = abs_file;
3135 #endif /* defined unix */
3139 clients->next = NULL;
3141 /* XXX: factor out initialising after the next stable release. */
3144 #elif defined(_WIN32)
3148 /* Prepare mutexes if supported and necessary. */
3149 initialize_mutexes();
3151 /* Enable logging until further notice. */
3152 init_log_module(Argv[0]);
3154 random_seed = (unsigned int)time(NULL);
3156 srandom(random_seed);
3157 #elif defined (_WIN32)
3159 * See pick_from_range() in miscutil.c for details.
3161 log_error(LOG_LEVEL_INFO,
3162 "No thread-safe PRNG implemented for your platform. "
3163 "Using weak \'randomization\' factor which will "
3164 "limit the already questionable usefulness of "
3165 "header-time-randomizing actions (disabled by default).");
3168 #endif /* ifdef HAVE_RANDOM */
3171 * Unix signal handling
3173 * Catch the abort, interrupt and terminate signals for a graceful exit
3174 * Catch the hangup signal so the errlog can be reopened.
3175 * Ignore the broken pipe signals (FIXME: Why?)
3177 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3180 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3181 const int ignored_signals[] = { SIGPIPE, 0 };
3183 for (idx = 0; catched_signals[idx] != 0; idx++)
3185 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3186 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3188 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3189 #endif /* ifdef sun */
3191 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3195 for (idx = 0; ignored_signals[idx] != 0; idx++)
3197 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3199 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3204 #else /* ifdef _WIN32 */
3205 # ifdef _WIN_CONSOLE
3207 * We *are* in a windows console app.
3208 * Print a verbose messages about FAQ's and such
3210 printf("%s", win32_blurb);
3211 # endif /* def _WIN_CONSOLE */
3212 #endif /* def _WIN32 */
3215 /* Initialize the CGI subsystem */
3216 cgi_init_error_messages();
3219 * If runnig on unix and without the --nodaemon
3220 * option, become a daemon. I.e. fork, detach
3221 * from tty and get process group leadership
3234 if ( pid < 0 ) /* error */
3239 else if ( pid != 0 ) /* parent */
3244 * must check for errors
3245 * child died due to missing files aso
3248 wpid = waitpid( pid, &status, WNOHANG );
3257 /* Should be more portable, but not as well tested */
3262 #else /* ndef __FreeBSD__ */
3264 #endif /* ndef __FreeBSD__ */
3265 fd = open("/dev/tty", O_RDONLY);
3268 /* no error check here */
3269 ioctl( fd, TIOCNOTTY,0 );
3274 * stderr (fd 2) will be closed later on, when the
3275 * log file has been parsed.
3282 } /* -END- if (!no_daemon) */
3285 * As soon as we have written the PID file, we can switch
3286 * to the user and group ID indicated by the --user option
3292 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3294 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3300 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3302 /* Read the time zone file from /etc before doing chroot. */
3304 if (NULL != pre_chroot_nslookup_to_load_resolver
3305 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3307 /* Initialize resolver library. */
3308 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3310 if (chroot(pw->pw_dir) < 0)
3312 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3316 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3319 if (setuid(pw->pw_uid))
3321 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3325 char putenv_dummy[64];
3327 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3328 if (putenv(putenv_dummy) != 0)
3330 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3333 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3334 if (putenv(putenv_dummy) != 0)
3336 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3342 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3345 #endif /* defined unix */
3348 /* This will be FALSE unless the command line specified --service
3352 /* Yup, so now we must attempt to establish a connection
3353 * with the service dispatcher. This will only work if this
3354 * process was launched by the service control manager to
3355 * actually run as a service. If this isn't the case, i've
3356 * known it take around 30 seconds or so for the call to return.
3359 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3360 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3362 /* Service has run, and at this point is now being stopped, so just return */
3367 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3369 /* An error occurred. Usually it's because --service was wrongly specified
3370 * and we were unable to connect to the Service Control Dispatcher because
3371 * it wasn't expecting us and is therefore not listening.
3373 * For now, just continue below to call the listen_loop function.
3376 #endif /* def _WIN32 */
3386 /*********************************************************************
3388 * Function : bind_port_helper
3390 * Description : Bind the listen port. Handles logging, and aborts
3394 * 1 : config = Privoxy configuration. Specifies port
3397 * Returns : Port that was opened.
3399 *********************************************************************/
3400 static jb_socket bind_port_helper(struct configuration_spec * config)
3405 if (config->haddr == NULL)
3407 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3412 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3413 config->hport, config->haddr);
3416 result = bind_port(config->haddr, config->hport, &bfd);
3423 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3424 "There may be another Privoxy or some other "
3425 "proxy running on port %d",
3426 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3427 config->hport, config->hport);
3430 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3431 "The hostname is not resolvable",
3432 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3435 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3436 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3439 /* shouldn't get here */
3440 return JB_INVALID_SOCKET;
3443 config->need_bind = 0;
3450 /* Without this simple workaround we get this compiler warning from _beginthread
3451 * warning C4028: formal parameter 1 different from declaration
3453 void w32_service_listen_loop(void *p)
3457 #endif /* def _WIN32 */
3460 /*********************************************************************
3462 * Function : listen_loop
3464 * Description : bind the listen port and enter a "FOREVER" listening loop.
3470 *********************************************************************/
3471 static void listen_loop(void)
3473 struct client_state *csp = NULL;
3475 struct configuration_spec * config;
3477 config = load_config();
3479 bfd = bind_port_helper(config);
3481 #ifdef FEATURE_GRACEFUL_TERMINATION
3482 while (!g_terminate)
3487 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3488 while (waitpid(-1, NULL, WNOHANG) > 0)
3490 /* zombie children */
3492 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3495 * Free data that was used by died threads
3501 * Re-open the errlog after HUP signal
3503 if (received_hup_signal)
3505 init_error_log(Argv[0], config->logfile);
3506 received_hup_signal = 0;
3511 #ifdef FEATURE_COOKIE_JAR
3513 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3514 * already open. I think unload_configfile() is not being run, which should do
3515 * this work. Until that can get resolved, we'll use this workaround.
3519 if (csp->config->jar)
3521 fclose(csp->config->jar);
3522 csp->config->jar = NULL;
3524 #endif /* FEATURE_COOKIE_JAR */
3525 #endif /* __OS2__ */
3527 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3529 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3533 csp->flags |= CSP_FLAG_ACTIVE;
3534 csp->sfd = JB_INVALID_SOCKET;
3536 csp->config = config = load_config();
3538 if ( config->need_bind )
3541 * Since we were listening to the "old port", we will not see
3542 * a "listen" param change until the next IJB request. So, at
3543 * least 1 more request must be made for us to find the new
3544 * setting. I am simply closing the old socket and binding the
3547 * Which-ever is correct, we will serve 1 more page via the
3548 * old settings. This should probably be a "show-proxy-args"
3549 * request. This should not be a so common of an operation
3550 * that this will hurt people's feelings.
3555 bfd = bind_port_helper(config);
3558 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3560 if (!accept_connection(csp, bfd))
3562 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3575 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3578 #ifdef FEATURE_TOGGLE
3579 if (global_toggle_state)
3580 #endif /* def FEATURE_TOGGLE */
3582 csp->flags |= CSP_FLAG_TOGGLED_ON;
3585 if (run_loader(csp))
3587 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3588 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3592 if (block_acl(NULL,csp))
3594 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3595 close_socket(csp->cfd);
3599 #endif /* def FEATURE_ACL */
3601 /* add it to the list of clients */
3602 csp->next = clients->next;
3603 clients->next = csp;
3605 if (config->multi_threaded)
3609 /* this is a switch () statment in the C preprocessor - ugh */
3610 #undef SELECTED_ONE_OPTION
3612 /* Use Pthreads in preference to native code */
3613 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3614 #define SELECTED_ONE_OPTION
3616 pthread_t the_thread;
3617 pthread_attr_t attrs;
3619 pthread_attr_init(&attrs);
3620 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3621 errno = pthread_create(&the_thread, &attrs,
3622 (void * (*)(void *))serve, csp);
3623 child_id = errno ? -1 : 0;
3624 pthread_attr_destroy(&attrs);
3628 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3629 #define SELECTED_ONE_OPTION
3630 child_id = _beginthread(
3631 (void (*)(void *))serve,
3636 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3637 #define SELECTED_ONE_OPTION
3638 child_id = _beginthread(
3639 (void(* _Optlink)(void*))serve,
3645 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3646 #define SELECTED_ONE_OPTION
3648 thread_id tid = spawn_thread
3649 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3651 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3653 child_id = (int) tid;
3662 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3663 #define SELECTED_ONE_OPTION
3664 csp->cfd = ReleaseSocket(csp->cfd, -1);
3667 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3668 NP_Output, Output(),
3669 NP_CloseOutput, FALSE,
3670 NP_Name, (ULONG)"privoxy child",
3674 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3675 NP_Output, Output(),
3676 NP_CloseOutput, FALSE,
3677 NP_Name, (ULONG)"privoxy child",
3678 NP_StackSize, 200*1024,
3684 ((struct Task *)child_id)->tc_UserData = csp;
3685 Signal((struct Task *)child_id, SIGF_SINGLE);
3690 #if !defined(SELECTED_ONE_OPTION)
3693 /* This block is only needed when using fork().
3694 * When using threads, the server thread was
3695 * created and run by the call to _beginthread().
3697 if (child_id == 0) /* child */
3700 #ifdef FEATURE_TOGGLE
3701 int inherited_toggle_state = global_toggle_state;
3702 #endif /* def FEATURE_TOGGLE */
3707 * If we've been toggled or we've blocked the request, tell Mom
3710 #ifdef FEATURE_TOGGLE
3711 if (inherited_toggle_state != global_toggle_state)
3713 rc |= RC_FLAG_TOGGLED;
3715 #endif /* def FEATURE_TOGGLE */
3717 #ifdef FEATURE_STATISTICS
3718 if (csp->flags & CSP_FLAG_REJECTED)
3720 rc |= RC_FLAG_BLOCKED;
3722 #endif /* ndef FEATURE_STATISTICS */
3726 else if (child_id > 0) /* parent */
3728 /* in a fork()'d environment, the parent's
3729 * copy of the client socket and the CSP
3733 #if !defined(_WIN32) && !defined(__CYGWIN__)
3735 wait( &child_status );
3738 * Evaluate child's return code: If the child has
3739 * - been toggled, toggle ourselves
3740 * - blocked its request, bump up the stats counter
3743 #ifdef FEATURE_TOGGLE
3744 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3746 global_toggle_state = !global_toggle_state;
3748 #endif /* def FEATURE_TOGGLE */
3750 #ifdef FEATURE_STATISTICS
3752 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3756 #endif /* def FEATURE_STATISTICS */
3758 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3759 close_socket(csp->cfd);
3760 csp->flags &= ~CSP_FLAG_ACTIVE;
3764 #undef SELECTED_ONE_OPTION
3765 /* end of cpp switch () */
3767 if (child_id < 0) /* failed */
3769 char buf[BUFFER_SIZE];
3771 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3773 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3775 write_socket(csp->cfd, buf, strlen(buf));
3776 close_socket(csp->cfd);
3777 csp->flags &= ~CSP_FLAG_ACTIVE;
3788 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3790 /* Clean up. Aim: free all memory (no leaks) */
3791 #ifdef FEATURE_GRACEFUL_TERMINATION
3793 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3795 unload_current_config_file();
3796 unload_current_actions_file();
3797 unload_current_re_filterfile();
3798 #ifdef FEATURE_TRUST
3799 unload_current_trust_file();
3802 if (config->multi_threaded)
3809 } while ((clients->next != NULL) && (--i > 0));
3813 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3824 #ifdef FEATURE_COOKIE_JAR
3825 if (NULL != config->jar)
3827 fclose(config->jar);
3831 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3832 /* Cleanup - remove taskbar icon etc. */
3837 #endif /* FEATURE_GRACEFUL_TERMINATION */