1 const char jcc_rcs[] = "$Id: jcc.c,v 1.180 2008/05/21 15:26:32 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 1.180 2008/05/21 15:26:32 fabiankeil
37 * - Mark csp as immutable for send_crunch_response().
38 * - Fix comment spelling.
40 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
41 * Factor update_server_headers() out of sed(), ditch the
42 * first_run hack and make server_patterns_light static.
44 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
45 * Don't provide get_header() with the whole client state
46 * structure when it only needs access to csp->iob.
48 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
49 * Make the "read the rest of the headers" loop a bit more readable.
51 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
52 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
53 * - Remove useless comment.
55 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
56 * Fix comment grammar.
58 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
59 * Remove the pointless buffer in client_protocol_is_unsupported().
61 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
62 * Least-effort fix for bug #1821930 (reported by Lee):
63 * If the response doesn't look like HTTP,
64 * tell the client and log the problem.
66 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
67 * Don't pass the whole csp structure to flush_socket()
68 * when it only needs a file descriptor and a buffer.
70 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
71 * Remove kill-popups action.
73 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
74 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
76 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
77 * Remove the treat-forbidden-connects-like-blocks action. We now
78 * use the "blocked" page for forbidden CONNECT requests by default.
80 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
81 * Also use shiny new connect_port_is_forbidden() in jcc.c.
83 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
84 * Rename url_actions() to get_url_actions() and let it
85 * use the standard parameter ordering.
87 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
88 * Let forward_url() use the standard parameter ordering
89 * and mark its second parameter immutable.
91 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
92 * Remove the "Listening ... for local connections only" log message.
93 * Whether or not remote connections are able to reach Privoxy is up
94 * to the operating system.
96 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
97 * Prevent the log messages for CONNECT requests to unacceptable
98 * ports from printing the limit-connect argument as [null] if
99 * limit-connect hasn't been explicitly enabled.
101 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
102 * Make sure all console-mode apps get a usage() instance
104 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
105 * Reword NO_SERVER_DATA_RESPONSE to make it harder
106 * to misunderstand what the message is all about.
108 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
109 * Unbreak trustfile which previously didn't work without
110 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
112 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
113 * Plug memory leak. Spotted by Valgrind, triggered by
114 * Privoxy-Regression-Test feeding proxyfuzz.py.
116 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
117 * In the HTTP snipplets, refer to the client as client.
119 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
120 * Emit a log message when activating the MS IIS5 hack.
122 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
123 * Log the "weak randomization factor" warning only
124 * once for mingw32 and provide some more details.
126 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
127 * Initialize log module after initializing mutexes, future
128 * deadlocks in that code should now work cross-platform.
130 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
131 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
132 * Reported by Bert van Leeuwen in BR#1818808.
134 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
135 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
137 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
138 * When in daemon mode, close stderr after the configuration file has been
139 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
141 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
142 * - Fix a crash when parsing invalid requests whose first header
143 * is rejected by get_header(). Regression (re?)introduced
144 * in r1.143 by yours truly.
145 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
146 * client_cookie_adder() to make sure send-vanilla-wafer can be
147 * controlled through tags (and thus regression-tested).
149 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
150 * - Move get_filter_function() from jcc.c to filters.c
151 * so the filter functions can be static.
152 * - Don't bother filtering body-less responses.
154 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
155 * Execute content filters through execute_content_filter().
157 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
158 * Initialize req to NULL to make sure it's defined if the
159 * first read_socket() call fails. Reported by icmp30.
161 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
162 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
163 * extensive comments moved to user manual.
165 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
166 * Don't crash if a broken header filter wiped out the request line.
168 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
169 * Fix byte_count calculation in case of flushes
170 * and don't parse the server headers a second time.
172 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
173 * - If there's a connection problem after we already forwarded
174 * parts of the original content, just hang up. Fixes BR#1776724.
175 * - Fix warnings about unused code on mingw32.
176 * - In case of flushes, calculate the byte count
177 * less incorrectly (I think).
179 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
180 * Add some more prototypes for static functions.
182 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
183 * Comment out request_contains_null_bytes() until it's used again.
185 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
186 * #1763173 from Stefan Huehner: s@const static@static const@
187 * and declare some more functions static.
189 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
190 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
191 * and explain why they were rejected in the first place.
192 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
193 * CONNECT requests. The request line was missing.
194 * - Add two more XXX reminders as we don't have enough already.
196 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
197 * As Hal noticed, checking dispatch_cgi() as the last cruncher
198 * looks like a bug if CGI requests are blocked unintentionally,
199 * so don't do it unless the user enabled the new config option
200 * "allow-cgi-request-crunching".
202 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
203 * - Allow to rewrite the request destination behind the client's back.
204 * - Turn the weird-looking unconditional for loop that
205 * reads the client request into a conditional while loop.
206 * Move the stuff that only runs once out of the loop.
207 * - Move parts of chat(), server_content_type() and the
208 * necessary stuff to fix BR#1750917 into get_filter_function().
210 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
211 * Temporary workaround for BR#1730105.
213 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
214 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
215 * deadlocks and crashes on OpenBSD and possibly other OS with neither
216 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
217 * Thanks to Ralf Horstmann for report and solution.
219 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
220 * Add forward-override{} to change the forwarding settings through
221 * action sections. This is mainly interesting to forward different
222 * clients differently (for example based on User-Agent or request
225 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
226 * - Let usage() mention the --chroot parameter.
227 * - Use read_socket() consistently and always leave
228 * the last buffer byte alone, even in cases where
229 * null termination (currently) doesn't matter.
231 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
232 * - Fix config file loading on Unix if no config file is specified.
233 * Since r1.97 Privoxy would always interpret the last argument as
234 * config file, even if it's a valid command line option.
235 * - Abort in case of unrecognized command line options. Closes #1719696.
236 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
237 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
239 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
240 * - Don't rerun crunchers that only depend on the request URL.
241 * - Don't count redirects and CGI requests as "blocked requests".
243 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
244 * Support crunching based on tags created by server-header taggers.
246 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
247 * Make HTTP snippets static (again). Add a Content-Type for those
248 * with content so the browser doesn't guess it based on the URL.
250 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
251 * Move crunching and request line rebuilding out of chat().
253 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
254 * Introduce tags as alternative way to specify which
255 * actions apply to a request. At the moment tags can be
256 * created based on client and server headers.
258 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
259 * Don't CLF-log CONNECT requests twice.
261 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
262 * Log the source address for ACL-related connection drops.
264 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
265 * New config option: enforce-blocks.
267 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
268 * - Move null byte check into separate function.
269 * - Don't confuse the client with error pages
270 * if a CONNECT request was already confirmed.
272 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
273 * Speed up NULL byte escaping and only log the complete
274 * NULL byte requests with header debugging enabled.
276 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
277 * Answer requests that contain NULL bytes with
278 * a custom response instead of waiting for more
279 * data until the client eventually hangs up.
281 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
282 * - Move delivery and logging of crunched responses
283 * from chat() into send_crunch_response().
284 * - Display the reason for generating http_responses.
285 * - Log the content length for LOG_LEVEL_CLF correctly
286 * (still incorrect for some fixed responses).
287 * - Reword an incorrect comment about
288 * treat-forbidden-connects-like-blocks violating
290 * - Add some log messages.
292 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
293 * Move mutex initialization into separate
294 * function and exit in case of errors.
296 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
297 * - Start to reduce chat()'s line count and move
298 * parts of it into separate functions.
299 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
301 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
302 * - Add Proxy-Agent header to HTTP snippets that are
303 * supposed to reach HTTP clients only.
304 * - Made a few CONNECT log messages more descriptive.
305 * - Catch completely empty server responses (as seen
306 * with Tor's fake ".noconnect" top level domain).
307 * - Use shiny new "forwarding-failed" template for socks errors.
309 * Revision 1.118 2007/01/07 07:43:43 joergs
310 * AmigaOS4 support added.
312 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
313 * Added config option accept-intercepted-requests
314 * and disabled it by default.
316 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
317 * Reverted parts of my last commit
318 * to keep error handling working.
320 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
321 * Fixed gcc43 conversion warnings.
323 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
324 * Fix -pedantic ISO C warning about converting
325 * from function pointer to object pointer.
327 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
328 * Silence compiler warning I introduced with my last commit.
330 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
331 * Mutex protect rand() if POSIX threading
332 * is used, warn the user if that's not possible
333 * and stop using it on _WIN32 where it could
336 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
337 * Don't prevent core dumps by catching SIGABRT.
338 * It's rude and makes debugging unreasonable painful.
340 * Revision 1.110 2006/12/13 14:52:53 etresoft
341 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
343 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
344 * Privoxy is now able to run as intercepting
345 * proxy in combination with any packet filter
346 * that does the port redirection. The destination
347 * is extracted from the "Host:" header which
348 * should be available for nearly all requests.
350 * Moved HTTP snipplets into jcc.c.
351 * Added error message for gopher proxy requests.
353 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
354 * Only unlink the pidfile if it's actually used.
356 * Change order of interception checks to make
357 * it possible to block or redirect requests for
360 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
361 * Make pthread mutex locking more generic. Instead of
362 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
363 * and use mutex locking unless there is an _r function
364 * available. Better safe than sorry.
366 * Fixes "./configure --disable-pthread" and should result
367 * in less threading-related problems on pthread-using platforms,
368 * but it still doesn't fix BR#1122404.
370 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
371 * Move pthread.h inclusion from jcc.c to jcc.h.
372 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
374 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
375 * Don't exit after receiving the second SIGHUP on Solaris.
377 * Fixes BR 1052235, but the same problem may exist on other
378 * systems. Once 3.0.6 is out we should use sigset()
379 * where available and see if it breaks anything.
381 * Revision 1.104 2006/09/23 13:26:38 roro
382 * Replace TABs by spaces in source code.
384 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
385 * Fix +redirect{}. Didn't work with -fast-redirects.
387 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
388 * Respond with 400 and a short text message
389 * if the client tries to use Privoxy as FTP proxy.
391 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
392 * Make number of retries in case of forwarded-connect problems
393 * a config file option (forwarded-connect-retries) and use 0 as
396 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
397 * Set random(3) seed.
399 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
400 * Follow the OpenBSD port's lead and protect the resolve
401 * functions on OpenBSD as well.
403 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
404 * --user fix. Only use the user as group if no group is specified.
405 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
407 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
408 * Windows service (re-)integration
412 * --install[:service_name]
413 * --uninstall[:service_name]
416 * They work as follows:
417 * --install will create a service for you and then terminate.
418 * By default the service name will be "privoxy" (without the quotes).
419 * However you can run multiple services if you wish, just by adding
420 * a colon and then a name (no spaces).
422 * --uninstall follows the exact same rules a --install.
424 * --service is used when the program is executed by the service
425 * control manager, and in normal circumstances would never be
426 * used as a command line argument.
428 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
429 * Windows service integration
431 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
432 * Incorporate Fabian Keil's patch work:
433 http://www.fabiankeil.de/sourcecode/privoxy/
435 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
436 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
437 * with what was really the latest development (the v_3_0_branch branch)
439 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
440 * Thanks to Jindrich Makovicka for a race condition fix for the log
441 * file. The race condition remains for non-pthread implementations.
442 * Reference patch #1175720.
444 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
445 * Add the ability to check jpeg images for invalid
446 * lengths of comment blocks. Defensive strategy
447 * against the exploit:
448 * Microsoft Security Bulletin MS04-028
449 * Buffer Overrun in JPEG Processing (GDI+) Could
450 * Allow Code Execution (833987)
451 * Enabled with +inspect-jpegs in actions files.
453 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
454 * - Fixed usage info for non-unix platforms
455 * - Fixed small cmdline parsing bug
457 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
458 * Diagnostics: Now preserve the returncode of pthread_create
459 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
461 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
462 * No longer ignore SIGCHLD. Fixes bug #769381
464 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
465 * Close jarfile on graceful exit, remove stray line
467 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
468 * Cosmetics: Killed a warning, a typo and an allocation left at exit
470 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
471 * No longer rely on non-POSIX.1 extensions of getcwd().
474 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
475 * Replaced setenv() by posix-compliant putenv()
476 * Thanks to Neil McCalden (nmcc AT users.sf.net).
478 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
479 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
481 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
482 * Clean-up and extension of improvements for forked mode:
483 * - Child's return code now consists of flags RC_FLAG_*
484 * - Reporting toggle to parent now properly #ifdef'ed
485 * - Children now report blocking to parent. This enables
486 * statistics in forked mode
488 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
489 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
490 * Children now report having been toggled through _exit(17), parents
491 * watch for that code and toggle themselves if found.
493 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
494 * Wrapping all *_r functions (the non-_r versions of them) with
495 * mutex semaphores for OSX. Hopefully this will take care of all
496 * of those pesky crash reports.
498 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
499 * Fixed two mostly harmless mem leaks
501 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
502 * Fix: Head of global clients list now initialized to NULL
504 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
505 * Added basic support for OPTIONS and TRACE HTTP methods:
506 * - New interceptor direct_response() added in chat().
507 * - sed() moved to earlier in the process, so that the
508 * Host: header is evaluated before actions and forwarding
511 * Revision 1.92 2002/05/08 16:00:46 oes
512 * Chat's buffer handling:
513 * - Fixed bug with unchecked out-of-mem conditions
514 * while reading client request & server headers
515 * - No longer predict if the buffer limit will be exceeded
516 * in the next read -- check add_to_iob's new
517 * return code. If buffer couldn't be extended
518 * (policy or out-of-mem) while
519 * - reading from client: abort
520 * - reading server headers: send error page
521 * - buffering server body for filter: flush,
522 * and if that fails: send error page
524 * Revision 1.91 2002/04/08 20:35:58 swa
527 * Revision 1.90 2002/04/02 14:57:28 oes
528 * Made sending wafers independent of FEATURE_COOKIE_JAR
530 * Revision 1.89 2002/03/31 17:18:59 jongfoster
531 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
533 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
534 * More compiler warning message maintenance
536 * Revision 1.87 2002/03/26 22:29:54 swa
537 * we have a new homepage!
539 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
540 * Workaround for closing the jarfile before load_config() comes around again
542 * Revision 1.85 2002/03/24 15:23:33 jongfoster
545 * Revision 1.84 2002/03/24 13:25:43 swa
546 * name change related issues
548 * Revision 1.83 2002/03/16 23:54:06 jongfoster
549 * Adding graceful termination feature, to help look for memory leaks.
550 * If you enable this (which, by design, has to be done by hand
551 * editing config.h) and then go to http://i.j.b/die, then the program
552 * will exit cleanly after the *next* request. It should free all the
553 * memory that was used.
555 * Revision 1.82 2002/03/13 00:27:05 jongfoster
558 * Revision 1.81 2002/03/12 01:42:50 oes
559 * Introduced modular filters
561 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
562 * OS/2 port maintenance:
563 * - Fixed EMX build - it had decayed a little
564 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
565 * substituted a memset for now.
567 * Revision 1.79 2002/03/09 20:03:52 jongfoster
568 * - Making various functions return int rather than size_t.
569 * (Undoing a recent change). Since size_t is unsigned on
570 * Windows, functions like read_socket that return -1 on
571 * error cannot return a size_t.
573 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
574 * crashes, and also frequently caused JB to jump to 100%
575 * CPU and stay there. (Because it thought it had just
576 * read ((unsigned)-1) == 4Gb of data...)
578 * - The signature of write_socket has changed, it now simply
579 * returns success=0/failure=nonzero.
581 * - Trying to get rid of a few warnings --with-debug on
582 * Windows, I've introduced a new type "jb_socket". This is
583 * used for the socket file descriptors. On Windows, this
584 * is SOCKET (a typedef for unsigned). Everywhere else, it's
585 * an int. The error value can't be -1 any more, so it's
586 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
587 * Windows it maps to the #define INVALID_SOCKET.)
589 * - The signature of bind_port has changed.
591 * Revision 1.78 2002/03/08 21:35:04 oes
592 * Added optional group supplement to --user option. Will now use default group of user if no group given
594 * Revision 1.77 2002/03/07 03:52:06 oes
595 * - Fixed compiler warnings etc
596 * - Improved handling of failed DNS lookups
598 * Revision 1.76 2002/03/06 22:54:35 jongfoster
599 * Automated function-comment nitpicking.
601 * Revision 1.75 2002/03/06 10:02:19 oes
602 * Fixed stupid bug when --user was not given
604 * Revision 1.74 2002/03/06 00:49:31 jongfoster
605 * Fixing warning on Windows
606 * Making #ifdefs that refer to the same variable consistently
607 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
609 * Revision 1.73 2002/03/05 23:57:30 hal9
610 * Stray character 's' on line 1618 was breaking build.
612 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
613 * - Re-enable OS/2 building after new parms were added
614 * - Fix false out of memory report when resolving CGI templates when no IP
615 * address is available of failed attempt (a la no such domain)
617 * Revision 1.71 2002/03/05 18:13:56 oes
618 * Added --user option
620 * Revision 1.70 2002/03/05 04:52:42 oes
621 * Deleted non-errlog debugging code
623 * Revision 1.69 2002/03/04 23:50:00 jongfoster
624 * Splitting off bind_port() call into bind_port_helper(), with
627 * Revision 1.68 2002/03/04 20:17:32 oes
630 * Revision 1.67 2002/03/04 18:18:57 oes
631 * - Removed _DEBUG mode
632 * - Cleand up cmdline parsing
633 * - Introduced --no-daemon, --pidfile options
634 * - Cleaned up signal handling:
635 * - Terminate cleanly on INT, TERM and ABRT
636 * - Schedule logfile for re-opening on HUP
637 * - Ignore CHLD and PIPE
638 * - Leave the rest with their default handlers
639 * - Uniform handler registration
640 * - Added usage() function
641 * - Played styleguide police
643 * Revision 1.66 2002/03/03 15:06:55 oes
644 * Re-enabled automatic config reloading
646 * Revision 1.65 2002/03/03 14:49:11 oes
647 * Fixed CLF logging: Now uses client's original HTTP request
649 * Revision 1.64 2002/03/03 09:18:03 joergs
650 * Made jumbjuster work on AmigaOS again.
652 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
653 * Clean up a little CRLF unpleasantness that suddenly appeared
655 * Revision 1.62 2002/02/20 23:17:23 jongfoster
656 * Detecting some out-of memory conditions and exiting with a log message.
658 * Revision 1.61 2002/01/17 21:01:52 jongfoster
659 * Moving all our URL and URL pattern parsing code to urlmatch.c.
661 * Revision 1.60 2001/12/30 14:07:32 steudten
662 * - Add signal handling (unix)
663 * - Add SIGHUP handler (unix)
664 * - Add creation of pidfile (unix)
665 * - Add action 'top' in rc file (RH)
666 * - Add entry 'SIGNALS' to manpage
667 * - Add exit message to logfile (unix)
669 * Revision 1.59 2001/12/13 14:07:18 oes
670 * Fixed Bug: 503 error page now sent OK
672 * Revision 1.58 2001/11/30 23:37:24 jongfoster
673 * Renaming the Win32 config file to config.txt - this is almost the
674 * same as the corresponding UNIX name "config"
676 * Revision 1.57 2001/11/16 00:47:43 jongfoster
677 * Changing the tty-disconnection code to use setsid().
679 * Revision 1.56 2001/11/13 20:20:54 jongfoster
680 * Tabs->spaces, fixing a bug with missing {} around an if()
682 * Revision 1.55 2001/11/13 20:14:53 jongfoster
683 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
685 * Revision 1.54 2001/11/07 00:03:14 steudten
686 * Give reliable return value if an error
687 * occurs not just 0 with new daemon mode.
689 * Revision 1.53 2001/11/05 21:41:43 steudten
690 * Add changes to be a real daemon just for unix os.
691 * (change cwd to /, detach from controlling tty, set
692 * process group and session leader to the own process.
694 * Add some fatal-error log message for failed malloc().
695 * Add '-d' if compiled with 'configure --with-debug' to
696 * enable debug output.
698 * Revision 1.52 2001/10/26 20:11:20 jongfoster
699 * Fixing type mismatch
701 * Revision 1.51 2001/10/26 17:38:28 oes
704 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
705 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
706 * threads to call select() simultaneously. So, it's time to do a real, live,
707 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
708 * (native). Both versions will work, but using __OS2__ offers multi-threading.
710 * Revision 1.49 2001/10/23 21:41:35 jongfoster
711 * Added call to initialize the (statically-allocated of course)
712 * "out of memory" CGI response.
714 * Revision 1.48 2001/10/10 19:56:46 jongfoster
715 * Moving some code that wasn't cookie-related out of an #ifdef
718 * Revision 1.47 2001/10/10 16:44:36 oes
719 * Added CONNECT destination port limitation check
721 * Revision 1.46 2001/10/08 15:17:41 oes
722 * Re-enabled SSL forwarding
724 * Revision 1.45 2001/10/07 15:42:11 oes
725 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
727 * Moved downgrading of the HTTP version from parse_http_request to
728 * chat(), since we can't decide if it is necessary before we have
729 * determined the actions for the URL. The HTTP command is now
730 * *always* re-built so the repairs need no longer be special-cased.
732 * filter_popups now gets a csp pointer so it can raise the new
733 * CSP_FLAG_MODIFIED flag.
737 * Added configurable size limit for the IOB. If the IOB grows so
738 * large that the next read would exceed the limit, the header
739 * is generated, and the header & unfiltered buffer are flushed
740 * to the client. Chat then continues in non-buffering,
741 * non-filtering body mode.
743 * Revision 1.44 2001/10/02 18:13:57 oes
746 * Revision 1.43 2001/10/02 15:32:13 oes
747 * Moved generation of hdr
749 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
750 * Cleaning up 2 compiler warnings on OS/2.
752 * Revision 1.41 2001/09/16 17:05:14 jongfoster
753 * Removing unused #include showarg.h
755 * Revision 1.40 2001/09/16 15:41:45 jongfoster
756 * Fixing signed/unsigned comparison warning.
758 * Revision 1.39 2001/09/16 13:21:27 jongfoster
759 * Changes to use new list functions.
761 * Revision 1.38 2001/09/16 13:01:46 jongfoster
762 * Removing redundant function call that zeroed zalloc()'d memory.
764 * Revision 1.37 2001/09/10 11:12:24 oes
765 * Deleted unused variable
767 * Revision 1.36 2001/09/10 10:56:15 oes
768 * Silenced compiler warnings
770 * Revision 1.35 2001/07/31 14:44:22 oes
771 * Deleted unused size parameter from filter_popups()
773 * Revision 1.34 2001/07/30 22:08:36 jongfoster
774 * Tidying up #defines:
775 * - All feature #defines are now of the form FEATURE_xxx
776 * - Permanently turned off WIN_GUI_EDIT
777 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
779 * Revision 1.33 2001/07/29 19:32:00 jongfoster
780 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
782 * Revision 1.32 2001/07/29 18:47:05 jongfoster
783 * Adding missing #include "loadcfg.h"
785 * Revision 1.31 2001/07/29 12:17:48 oes
786 * Applied pthread fix by Paul Lieverse
788 * Revision 1.30 2001/07/25 22:57:13 jongfoster
789 * __BEOS__ no longer overrides FEATURE_PTHREAD.
790 * This is because FEATURE_PTHREAD will soon be widely used, so I
791 * want to keep it simple.
793 * Revision 1.29 2001/07/24 12:47:06 oes
794 * Applied BeOS support update by Eugenia
796 * Revision 1.28 2001/07/23 13:26:12 oes
797 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
799 * Revision 1.27 2001/07/19 19:09:47 haroon
800 * - Added code to take care of the situation where while processing the first
801 * server response (which includes the server header), after finding the end
802 * of the headers we were not looking past the end of the headers for
803 * content modification. I enabled it for filter_popups.
804 * Someone else should look to see if other similar operations should be
805 * done to the discarded portion of the buffer.
807 * Note 2001/07/20: No, the other content modification mechanisms will process
808 * the whole iob later anyway. --oes
810 * Revision 1.26 2001/07/18 12:31:36 oes
813 * Revision 1.25 2001/07/15 19:43:49 jongfoster
814 * Supports POSIX threads.
815 * Also removed some unused #includes.
817 * Revision 1.24 2001/07/13 14:00:40 oes
818 * - Generic content modification scheme:
819 * Each feature has its own applicability flag that is set
820 * from csp->action->flags.
821 * Replaced the "filtering" int flag , by a function pointer
822 * "content_filter" to the function that will do the content
823 * modification. If it is != NULL, the document will be buffered
824 * and processed through *content_filter, which must set
825 * csp->content_length and return a modified copy of the body
826 * or return NULL (on failiure).
827 * - Changed csp->is_text to the more generic bitmap csp->content_type
828 * which can currently take the valued CT_TEXT or CT_GIF
830 * - Removed all #ifdef PCRS
832 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
833 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
834 * the `block_acl' call. This prevents a compilation error when the user
835 * does not wish to use the "ACL" feature.
837 * Revision 1.22 2001/06/29 21:45:41 oes
838 * Indentation, CRLF->LF, Tab-> Space
840 * Revision 1.21 2001/06/29 13:29:36 oes
841 * - Cleaned up, improved comments
842 * - Unified all possible interceptors (CGI,
843 * block, trust, fast_redirect) in one
844 * place, with one (CGI) answer generation
845 * mechansim. Much clearer now.
846 * - Removed the GIF image generation, which
847 * is now done in filters.c:block_url()
848 * - Made error conditions like domain lookup
849 * failiure or (various) problems while talking
850 * to the server use cgi.c:error_response()
851 * instead of generating HTML/HTTP in chat() (yuck!)
852 * - Removed logentry from cancelled commit
854 * Revision 1.20 2001/06/09 10:55:28 jongfoster
855 * Changing BUFSIZ ==> BUFFER_SIZE
857 * Revision 1.19 2001/06/07 23:12:52 jongfoster
858 * Replacing function pointer in struct gateway with a directly
859 * called function forwarded_connect().
860 * Replacing struct gateway with struct forward_spec
862 * Revision 1.18 2001/06/03 19:12:16 oes
863 * introduced new cgi handling
865 * Revision 1.17 2001/06/01 20:07:23 jongfoster
866 * Now uses action +image-blocker{} rather than config->tinygif
868 * Revision 1.16 2001/06/01 18:49:17 jongfoster
869 * Replaced "list_share" with "list" - the tiny memory gain was not
870 * worth the extra complexity.
872 * Revision 1.15 2001/05/31 21:24:47 jongfoster
873 * Changed "permission" to "action" throughout.
874 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
875 * Moved vanilla wafer check into chat(), since we must now
876 * decide whether or not to add it based on the URL.
878 * Revision 1.14 2001/05/29 20:14:01 joergs
879 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
882 * Revision 1.13 2001/05/29 09:50:24 jongfoster
883 * Unified blocklist/imagelist/permissionslist.
884 * File format is still under discussion, but the internal changes
887 * Also modified interceptor behaviour:
888 * - We now intercept all URLs beginning with one of the following
889 * prefixes (and *only* these prefixes):
891 * * http://ijbswa.sf.net/config/
892 * * http://ijbswa.sourceforge.net/config/
893 * - New interceptors "home page" - go to http://i.j.b/ to see it.
894 * - Internal changes so that intercepted and fast redirect pages
895 * are not replaced with an image.
896 * - Interceptors now have the option to send a binary page direct
897 * to the client. (i.e. ijb-send-banner uses this)
898 * - Implemented show-url-info interceptor. (Which is why I needed
899 * the above interceptors changes - a typical URL is
900 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
901 * The previous mechanism would not have intercepted that, and
902 * if it had been intercepted then it then it would have replaced
905 * Revision 1.12 2001/05/27 22:17:04 oes
907 * - re_process_buffer no longer writes the modified buffer
908 * to the client, which was very ugly. It now returns the
909 * buffer, which it is then written by chat.
911 * - content_length now adjusts the Content-Length: header
912 * for modified documents rather than crunch()ing it.
913 * (Length info in csp->content_length, which is 0 for
914 * unmodified documents)
916 * - For this to work, sed() is called twice when filtering.
918 * Revision 1.11 2001/05/26 17:27:53 jongfoster
919 * Added support for CLF and fixed LOG_LEVEL_LOG.
920 * Also did CRLF->LF fix of my previous patch.
922 * Revision 1.10 2001/05/26 15:26:15 jongfoster
923 * ACL feature now provides more security by immediately dropping
924 * connections from untrusted hosts.
926 * Revision 1.9 2001/05/26 00:28:36 jongfoster
927 * Automatic reloading of config file.
928 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
929 * Most of the global variables have been moved to a new
930 * struct configuration_spec, accessed through csp->config->globalname
931 * Most of the globals remaining are used by the Win32 GUI.
933 * Revision 1.8 2001/05/25 22:43:18 jongfoster
934 * Fixing minor memory leak and buffer overflow.
936 * Revision 1.7 2001/05/25 22:34:30 jongfoster
939 * Revision 1.6 2001/05/23 00:13:58 joergs
940 * AmigaOS support fixed.
942 * Revision 1.5 2001/05/22 18:46:04 oes
944 * - Enabled filtering banners by size rather than URL
945 * by adding patterns that replace all standard banner
946 * sizes with the "Junkbuster" gif to the re_filterfile
948 * - Enabled filtering WebBugs by providing a pattern
949 * which kills all 1x1 images
951 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
952 * which is selected by the (nonstandard and therefore
953 * capital) letter 'U' in the option string.
954 * It causes the quantifiers to be ungreedy by default.
955 * Appending a ? turns back to greedy (!).
957 * - Added a new interceptor ijb-send-banner, which
958 * sends back the "Junkbuster" gif. Without imagelist or
959 * MSIE detection support, or if tinygif = 1, or the
960 * URL isn't recognized as an imageurl, a lame HTML
961 * explanation is sent instead.
963 * - Added new feature, which permits blocking remote
964 * script redirects and firing back a local redirect
966 * The feature is conditionally compiled, i.e. it
967 * can be disabled with --disable-fast-redirects,
968 * plus it must be activated by a "fast-redirects"
969 * line in the config file, has its own log level
970 * and of course wants to be displayed by show-proxy-args
971 * Note: Boy, all the #ifdefs in 1001 locations and
972 * all the fumbling with configure.in and acconfig.h
973 * were *way* more work than the feature itself :-(
975 * - Because a generic redirect template was needed for
976 * this, tinygif = 3 now uses the same.
978 * - Moved GIFs, and other static HTTP response templates
983 * - Removed some >400 CRs again (Jon, you really worked
986 * Revision 1.4 2001/05/21 19:34:01 jongfoster
987 * Made failure to bind() a fatal error.
989 * Revision 1.3 2001/05/20 01:21:20 jongfoster
990 * Version 2.9.4 checkin.
991 * - Merged popupfile and cookiefile, and added control over PCRS
992 * filtering, in new "permissionsfile".
993 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
994 * file error you now get a message box (in the Win32 GUI) rather
995 * than the program exiting with no explanation.
996 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
998 * - Removed tabs from "config"
999 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1000 * - Bumped up version number.
1002 * Revision 1.2 2001/05/17 22:34:44 oes
1003 * - Added hint on GIF char array generation to jcc.c
1004 * - Cleaned CRLF's from the sources and related files
1005 * - Repaired logging for REF and FRC
1007 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1008 * Initial import of version 2.9.3 source tree
1011 *********************************************************************/
1017 #include <sys/types.h>
1026 # ifndef FEATURE_PTHREAD
1030 # include <windows.h>
1031 # include <process.h>
1032 # endif /* ndef FEATURE_PTHREAD */
1035 # ifndef _WIN_CONSOLE
1036 # include "w32log.h"
1037 # endif /* ndef _WIN_CONSOLE */
1038 # include "w32svrapi.h"
1040 #else /* ifndef _WIN32 */
1042 # if !defined (__OS2__)
1043 # include <unistd.h>
1044 # include <sys/wait.h>
1045 # endif /* ndef __OS2__ */
1046 # include <sys/time.h>
1047 # include <sys/stat.h>
1048 # include <sys/ioctl.h>
1051 #include <sys/termios.h>
1059 # include <signal.h>
1062 # include <socket.h> /* BeOS has select() for sockets only. */
1063 # include <OS.h> /* declarations for threads and stuff. */
1066 # if defined(__EMX__) || defined(__OS2__)
1067 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1072 #define bzero(B,N) memset(B,0x00,n)
1076 # include <select.h>
1081 #include "project.h"
1084 #include "filters.h"
1085 #include "loaders.h"
1086 #include "parsers.h"
1087 #include "miscutil.h"
1089 #include "jbsockets.h"
1090 #include "gateway.h"
1091 #include "actions.h"
1093 #include "loadcfg.h"
1094 #include "urlmatch.h"
1096 const char jcc_h_rcs[] = JCC_H_VERSION;
1097 const char project_h_rcs[] = PROJECT_H_VERSION;
1100 struct client_state clients[1];
1101 struct file_list files[1];
1103 #ifdef FEATURE_STATISTICS
1104 int urls_read = 0; /* total nr of urls read inc rejected */
1105 int urls_rejected = 0; /* total nr of urls rejected */
1106 #endif /* def FEATURE_STATISTICS */
1108 #ifdef FEATURE_GRACEFUL_TERMINATION
1109 int g_terminate = 0;
1112 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1113 static void sig_handler(int the_signal);
1115 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1116 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1117 static jb_err get_server_headers(struct client_state *csp);
1118 static const char *crunch_reason(const struct http_response *rsp);
1119 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1121 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1123 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1124 static jb_err change_request_destination(struct client_state *csp);
1125 static void chat(struct client_state *csp);
1126 static void serve(struct client_state *csp);
1127 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1128 static void usage(const char *myname);
1130 static void initialize_mutexes(void);
1131 static jb_socket bind_port_helper(struct configuration_spec *config);
1132 static void listen_loop(void);
1135 void serve(struct client_state *csp);
1136 #else /* ifndef AMIGA */
1137 static void serve(struct client_state *csp);
1138 #endif /* def AMIGA */
1141 static int32 server_thread(void *data);
1142 #endif /* def __BEOS__ */
1145 #define sleep(N) Sleep(((N) * 1000))
1149 #define sleep(N) DosSleep(((N) * 100))
1152 #ifdef FEATURE_PTHREAD
1153 pthread_mutex_t log_mutex;
1154 pthread_mutex_t log_init_mutex;
1156 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1157 pthread_mutex_t resolver_mutex;
1158 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1160 #ifndef HAVE_GMTIME_R
1161 pthread_mutex_t gmtime_mutex;
1162 #endif /* ndef HAVE_GMTIME_R */
1164 #ifndef HAVE_LOCALTIME_R
1165 pthread_mutex_t localtime_mutex;
1166 #endif /* ndef HAVE_GMTIME_R */
1169 pthread_mutex_t rand_mutex;
1170 #endif /* ndef HAVE_RANDOM */
1172 #endif /* FEATURE_PTHREAD */
1174 #if defined(unix) || defined(__EMX__)
1175 const char *basedir = NULL;
1176 const char *pidfile = NULL;
1177 int received_hup_signal = 0;
1178 #endif /* defined unix */
1180 /* HTTP snipplets. */
1181 static const char CSUCCEED[] =
1182 "HTTP/1.0 200 Connection established\r\n"
1183 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1185 static const char CHEADER[] =
1186 "HTTP/1.0 400 Invalid header received from client\r\n"
1187 "Proxy-Agent: Privoxy " VERSION "\r\n"
1188 "Content-Type: text/plain\r\n"
1189 "Connection: close\r\n\r\n"
1190 "Invalid header received from client.\r\n";
1192 static const char FTP_RESPONSE[] =
1193 "HTTP/1.0 400 Invalid request received from client\r\n"
1194 "Content-Type: text/plain\r\n"
1195 "Connection: close\r\n\r\n"
1196 "Invalid request. Privoxy doesn't support FTP.\r\n";
1198 static const char GOPHER_RESPONSE[] =
1199 "HTTP/1.0 400 Invalid request received from client\r\n"
1200 "Content-Type: text/plain\r\n"
1201 "Connection: close\r\n\r\n"
1202 "Invalid request. Privoxy doesn't support gopher.\r\n";
1204 /* XXX: should be a template */
1205 static const char MISSING_DESTINATION_RESPONSE[] =
1206 "HTTP/1.0 400 Bad request received from client\r\n"
1207 "Proxy-Agent: Privoxy " VERSION "\r\n"
1208 "Content-Type: text/plain\r\n"
1209 "Connection: close\r\n\r\n"
1210 "Bad request. Privoxy was unable to extract the destination.\r\n";
1212 /* XXX: should be a template */
1213 static const char NO_SERVER_DATA_RESPONSE[] =
1214 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1215 "Proxy-Agent: Privoxy " VERSION "\r\n"
1216 "Content-Type: text/plain\r\n"
1217 "Connection: close\r\n\r\n"
1218 "Empty server or forwarder response.\r\n"
1219 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1221 /* XXX: should be a template */
1222 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1223 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1224 "Proxy-Agent: Privoxy " VERSION "\r\n"
1225 "Content-Type: text/plain\r\n"
1226 "Connection: close\r\n\r\n"
1227 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1230 /* XXX: should be a template */
1231 static const char NULL_BYTE_RESPONSE[] =
1232 "HTTP/1.0 400 Bad request received from client\r\n"
1233 "Proxy-Agent: Privoxy " VERSION "\r\n"
1234 "Content-Type: text/plain\r\n"
1235 "Connection: close\r\n\r\n"
1236 "Bad request. Null byte(s) before end of request.\r\n";
1239 /* XXX: should be a template */
1240 static const char MESSED_UP_REQUEST_RESPONSE[] =
1241 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1242 "Proxy-Agent: Privoxy " VERSION "\r\n"
1243 "Content-Type: text/plain\r\n"
1244 "Connection: close\r\n\r\n"
1245 "Bad request. Messed up with header filters.\r\n";
1247 /* A function to crunch a response */
1248 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1250 /* Crunch function flags */
1251 #define CF_NO_FLAGS 0
1252 /* Cruncher applies to forced requests as well */
1253 #define CF_IGNORE_FORCE 1
1254 /* Crunched requests are counted for the block statistics */
1255 #define CF_COUNT_AS_REJECT 2
1257 /* A crunch function and its flags */
1260 const crunch_func_ptr cruncher;
1264 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1266 /* Complete list of cruncher functions */
1267 static const struct cruncher crunchers_all[] = {
1268 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1269 { block_url, CF_COUNT_AS_REJECT },
1270 #ifdef FEATURE_TRUST
1271 { trust_url, CF_COUNT_AS_REJECT },
1272 #endif /* def FEATURE_TRUST */
1273 { redirect_url, CF_NO_FLAGS },
1274 { dispatch_cgi, CF_IGNORE_FORCE},
1278 /* Light version, used after tags are applied */
1279 static const struct cruncher crunchers_light[] = {
1280 { block_url, CF_COUNT_AS_REJECT },
1281 { redirect_url, CF_NO_FLAGS },
1286 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1287 /*********************************************************************
1289 * Function : sig_handler
1291 * Description : Signal handler for different signals.
1292 * Exit gracefully on TERM and INT
1293 * or set a flag that will cause the errlog
1294 * to be reopened by the main thread on HUP.
1297 * 1 : the_signal = the signal cause this function to call
1301 *********************************************************************/
1302 static void sig_handler(int the_signal)
1308 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1319 received_hup_signal = 1;
1324 * We shouldn't be here, unless we catch signals
1325 * in main() that we can't handle here!
1327 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1335 /*********************************************************************
1337 * Function : client_protocol_is_unsupported
1339 * Description : Checks if the client used a known unsupported
1340 * protocol and deals with it by sending an error
1344 * 1 : csp = Current client state (buffers, headers, etc...)
1345 * 2 : req = the first request line send by the client
1347 * Returns : TRUE if an error response has been generated, or
1348 * FALSE if the request doesn't look invalid.
1350 *********************************************************************/
1351 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1354 * If it's a FTP or gopher request, we don't support it.
1356 * These checks are better than nothing, but they might
1357 * not work in all configurations and some clients might
1358 * have problems digesting the answer.
1360 * They should, however, never cause more problems than
1361 * Privoxy's old behaviour (returning the misleading HTML
1364 * "Could not resolve http://(ftp|gopher)://example.org").
1366 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1368 const char *response = NULL;
1369 const char *protocol = NULL;
1371 if (!strncmpic(req, "GET ftp://", 10))
1373 response = FTP_RESPONSE;
1378 response = GOPHER_RESPONSE;
1379 protocol = "GOPHER";
1381 log_error(LOG_LEVEL_ERROR,
1382 "%s tried to use Privoxy as %s proxy: %s",
1383 csp->ip_addr_str, protocol, req);
1384 log_error(LOG_LEVEL_CLF,
1385 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1387 write_socket(csp->cfd, response, strlen(response));
1396 /*********************************************************************
1398 * Function : get_request_destination_elsewhere
1400 * Description : If the client's request was redirected into
1401 * Privoxy without the client's knowledge,
1402 * the request line lacks the destination host.
1404 * This function tries to get it elsewhere,
1405 * provided accept-intercepted-requests is enabled.
1407 * "Elsewhere" currently only means "Host: header",
1408 * but in the future we may ask the redirecting
1409 * packet filter to look the destination up.
1411 * If the destination stays unknown, an error
1412 * response is send to the client and headers
1413 * are freed so that chat() can return directly.
1416 * 1 : csp = Current client state (buffers, headers, etc...)
1417 * 2 : headers = a header list
1419 * Returns : JB_ERR_OK if the destination is now known, or
1420 * JB_ERR_PARSE if it isn't.
1422 *********************************************************************/
1423 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1427 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1429 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1430 " Privoxy isn't configured to accept intercepted requests.",
1431 csp->ip_addr_str, csp->http->cmd);
1432 /* XXX: Use correct size */
1433 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1434 csp->ip_addr_str, csp->http->cmd);
1436 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1437 destroy_list(headers);
1439 return JB_ERR_PARSE;
1441 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1443 /* Split the domain we just got for pattern matching */
1444 init_domain_components(csp->http);
1450 /* We can't work without destination. Go spread the news.*/
1452 req = list_to_text(headers);
1454 /* XXX: Use correct size */
1455 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1456 csp->ip_addr_str, csp->http->cmd);
1457 log_error(LOG_LEVEL_ERROR,
1458 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1459 csp->ip_addr_str, csp->http->cmd, req);
1462 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1463 destroy_list(headers);
1465 return JB_ERR_PARSE;
1468 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1469 * to get the destination IP address, use it as host directly
1470 * or do a reverse DNS lookup first.
1475 /*********************************************************************
1477 * Function : get_server_headers
1479 * Description : Parses server headers in iob and fills them
1480 * into csp->headers so that they can later be
1484 * 1 : csp = Current client state (buffers, headers, etc...)
1486 * Returns : JB_ERR_OK if everything went fine, or
1487 * JB_ERR_PARSE if the headers were incomplete.
1489 *********************************************************************/
1490 static jb_err get_server_headers(struct client_state *csp)
1492 int continue_hack_in_da_house = 0;
1495 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1500 * continue hack in da house. Ignore the ending of
1501 * this head and continue enlisting header lines.
1502 * The reason is described below.
1504 enlist(csp->headers, "");
1505 continue_hack_in_da_house = 0;
1508 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1511 * It's a bodyless continue response, don't
1512 * stop header parsing after reaching its end.
1514 * As a result Privoxy will concatenate the
1515 * next response's head and parse and deliver
1516 * the headers as if they belonged to one request.
1518 * The client will separate them because of the
1519 * empty line between them.
1521 * XXX: What we're doing here is clearly against
1522 * the intended purpose of the continue header,
1523 * and under some conditions (HTTP/1.0 client request)
1524 * it's a standard violation.
1526 * Anyway, "sort of against the spec" is preferable
1527 * to "always getting confused by Continue responses"
1528 * (Privoxy's behaviour before this hack was added)
1530 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1531 continue_hack_in_da_house = 1;
1533 else if (*header == '\0')
1536 * If the header is empty, but the Continue hack
1537 * isn't active, we can assume that we reached the
1538 * end of the buffer before we hit the end of the
1541 * Inform the caller an let it decide how to handle it.
1543 return JB_ERR_PARSE;
1546 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1549 * XXX: Should we quit the request and return a
1550 * out of memory error page instead?
1552 log_error(LOG_LEVEL_ERROR,
1553 "Out of memory while enlisting server headers. %s lost.",
1563 /*********************************************************************
1565 * Function : crunch_reason
1567 * Description : Translates the crunch reason code into a string.
1570 * 1 : rsp = a http_response
1572 * Returns : A string with the crunch reason or an error description.
1574 *********************************************************************/
1575 static const char *crunch_reason(const struct http_response *rsp)
1577 char * reason = NULL;
1579 assert(rsp != NULL);
1582 return "Internal error while searching for crunch reason";
1585 switch (rsp->reason)
1587 case RSP_REASON_UNSUPPORTED:
1588 reason = "Unsupported HTTP feature";
1590 case RSP_REASON_BLOCKED:
1593 case RSP_REASON_UNTRUSTED:
1594 reason = "Untrusted";
1596 case RSP_REASON_REDIRECTED:
1597 reason = "Redirected";
1599 case RSP_REASON_CGI_CALL:
1600 reason = "CGI Call";
1602 case RSP_REASON_NO_SUCH_DOMAIN:
1603 reason = "DNS failure";
1605 case RSP_REASON_FORWARDING_FAILED:
1606 reason = "Forwarding failed";
1608 case RSP_REASON_CONNECT_FAILED:
1609 reason = "Connection failure";
1611 case RSP_REASON_OUT_OF_MEMORY:
1612 reason = "Out of memory (may mask other reasons)";
1615 reason = "No reason recorded";
1623 /*********************************************************************
1625 * Function : send_crunch_response
1627 * Description : Delivers already prepared response for
1628 * intercepted requests, logs the interception
1629 * and frees the response.
1632 * 1 : csp = Current client state (buffers, headers, etc...)
1633 * 1 : rsp = Fully prepared response. Will be freed on exit.
1635 * Returns : Nothing.
1637 *********************************************************************/
1638 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1640 const struct http_request *http = csp->http;
1641 char status_code[4];
1643 assert(rsp != NULL);
1644 assert(rsp->head != NULL);
1649 * Not supposed to happen. If it does
1650 * anyway, treat it as an unknown error.
1652 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1653 /* return code doesn't matter */
1658 /* If rsp is still NULL, we have serious internal problems. */
1659 log_error(LOG_LEVEL_FATAL,
1660 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1664 * Extract the status code from the actual head
1665 * that was send to the client. It is the only
1666 * way to get it right for all requests, including
1667 * the fixed ones for out-of-memory problems.
1669 * A head starts like this: 'HTTP/1.1 200...'
1673 status_code[0] = rsp->head[9];
1674 status_code[1] = rsp->head[10];
1675 status_code[2] = rsp->head[11];
1676 status_code[3] = '\0';
1678 /* Write the answer to the client */
1679 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1680 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1682 /* There is nothing we can do about it. */
1683 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1686 /* Log that the request was crunched and why. */
1687 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1688 http->hostport, http->path, crunch_reason(rsp));
1689 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1690 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1692 /* Clean up and return */
1693 if (cgi_error_memory() != rsp)
1695 free_http_response(rsp);
1702 /*********************************************************************
1704 * Function : request_contains_null_bytes
1706 * Description : Checks for NULL bytes in the request and sends
1707 * an error message to the client if any were found.
1709 * XXX: currently not used, see comment in chat().
1712 * 1 : csp = Current client state (buffers, headers, etc...)
1713 * 2 : buf = Data from the client's request to check.
1714 * 3 : len = The data length.
1716 * Returns : TRUE if the request contained one or more NULL bytes, or
1719 *********************************************************************/
1720 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1722 size_t c_len; /* Request lenght when treated as C string */
1724 c_len = strlen(buf);
1729 * Null byte(s) found. Log the request,
1730 * return an error response and hang up.
1732 size_t tmp_len = c_len;
1737 * Replace NULL byte(s) with '°' characters
1738 * so the request can be logged as string.
1739 * XXX: Is there a better replacement character?
1742 tmp_len += strlen(buf+tmp_len);
1743 } while (tmp_len < len);
1745 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1746 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1747 log_error(LOG_LEVEL_HEADER,
1748 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1750 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1752 /* XXX: Log correct size */
1753 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1763 /*********************************************************************
1765 * Function : crunch_response_triggered
1767 * Description : Checks if the request has to be crunched,
1768 * and delivers the crunch response if necessary.
1771 * 1 : csp = Current client state (buffers, headers, etc...)
1772 * 2 : crunchers = list of cruncher functions to run
1774 * Returns : TRUE if the request was answered with a crunch response
1777 *********************************************************************/
1778 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1780 struct http_response *rsp = NULL;
1781 const struct cruncher *c;
1784 * If CGI request crunching is disabled,
1785 * check the CGI dispatcher out of order to
1786 * prevent unintentional blocks or redirects.
1788 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1789 && (NULL != (rsp = dispatch_cgi(csp))))
1791 /* Deliver, log and free the interception response. */
1792 send_crunch_response(csp, rsp);
1796 for (c = crunchers; c->cruncher != NULL; c++)
1799 * Check the cruncher if either Privoxy is toggled
1800 * on and the request isn't forced, or if the cruncher
1801 * applies to forced requests as well.
1803 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1804 !(csp->flags & CSP_FLAG_FORCED)) ||
1805 (c->flags & CF_IGNORE_FORCE))
1807 rsp = c->cruncher(csp);
1810 /* Deliver, log and free the interception response. */
1811 send_crunch_response(csp, rsp);
1812 #ifdef FEATURE_STATISTICS
1813 if (c->flags & CF_COUNT_AS_REJECT)
1815 csp->flags |= CSP_FLAG_REJECTED;
1817 #endif /* def FEATURE_STATISTICS */
1828 /*********************************************************************
1830 * Function : build_request_line
1832 * Description : Builds the HTTP request line.
1834 * If a HTTP forwarder is used it expects the whole URL,
1835 * web servers only get the path.
1838 * 1 : csp = Current client state (buffers, headers, etc...)
1839 * 2 : fwd = The forwarding spec used for the request
1840 * XXX: Should use http->fwd instead.
1841 * 3 : request_line = The old request line which will be replaced.
1843 * Returns : Nothing. Terminates in case of memory problems.
1845 *********************************************************************/
1846 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1848 struct http_request *http = csp->http;
1850 assert(http->ssl == 0);
1853 * Downgrade http version from 1.1 to 1.0
1854 * if +downgrade action applies.
1856 if ( (csp->action->flags & ACTION_DOWNGRADE)
1857 && (!strcmpic(http->ver, "HTTP/1.1")))
1860 http->ver = strdup("HTTP/1.0");
1862 if (http->ver == NULL)
1864 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1869 * Rebuild the request line.
1871 freez(*request_line);
1872 *request_line = strdup(http->gpc);
1873 string_append(request_line, " ");
1875 if (fwd->forward_host)
1877 string_append(request_line, http->url);
1881 string_append(request_line, http->path);
1883 string_append(request_line, " ");
1884 string_append(request_line, http->ver);
1886 if (*request_line == NULL)
1888 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1890 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1894 /*********************************************************************
1896 * Function : change_request_destination
1898 * Description : Parse a (rewritten) request line and regenerate
1899 * the http request data.
1902 * 1 : csp = Current client state (buffers, headers, etc...)
1904 * Returns : Forwards the parse_http_request() return code.
1905 * Terminates in case of memory problems.
1907 *********************************************************************/
1908 static jb_err change_request_destination(struct client_state *csp)
1910 struct http_request *http = csp->http;
1913 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1914 free_http_request(http);
1915 err = parse_http_request(csp->headers->first->str, http, csp);
1916 if (JB_ERR_OK != err)
1918 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1919 jb_err_to_string(err));
1921 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1922 if (http->ocmd == NULL)
1924 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1931 /*********************************************************************
1935 * Description : Once a connection to the client has been accepted,
1936 * this function is called (via serve()) to handle the
1937 * main business of the communication. When this
1938 * function returns, the caller must close the client
1941 * FIXME: chat is nearly thousand lines long.
1945 * 1 : csp = Current client state (buffers, headers, etc...)
1947 * Returns : Nothing.
1949 *********************************************************************/
1950 static void chat(struct client_state *csp)
1952 char buf[BUFFER_SIZE];
1960 int ms_iis5_hack = 0;
1961 size_t byte_count = 0;
1962 int forwarded_connect_retries = 0;
1963 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1964 const struct forward_spec * fwd;
1965 struct http_request *http;
1966 int len; /* for buffer sizes (and negative error codes) */
1969 /* Function that does the content filtering for the current request */
1970 filter_function_ptr content_filter = NULL;
1972 /* Skeleton for HTTP response, if we should intercept the request */
1973 struct http_response *rsp;
1975 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1976 struct list header_list;
1977 struct list *headers = &header_list;
1981 memset(buf, 0, sizeof(buf));
1984 * Read the client's request. Note that since we're not using select() we
1985 * could get blocked here if a client connected, then didn't say anything!
1990 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1992 if (len <= 0) break; /* error! */
1995 * If there is no memory left for buffering the
1996 * request, there is nothing we can do but hang up
1998 if (add_to_iob(csp, buf, len))
2003 req = get_header(csp->iob);
2005 } while ((NULL != req) && ('\0' == *req));
2007 if ((NULL != req) && ('\0' != *req))
2009 /* Request received. Validate and parse it. */
2013 * XXX: Temporary disabled to prevent problems
2014 * with POST requests whose bodies are allowed to
2015 * contain NULL bytes. BR#1730105.
2017 * The main purpose of this check is to properly
2018 * log stuff like BitTorrent traffic and other junk
2019 * that hits public proxies. It's not required for
2020 * Privoxy to functions as those requests are discarded
2023 * It probably should be rewritten to only check
2024 * the head of the request. Another option would
2025 * be to let all POST requests pass, although that
2026 * may not be good enough.
2028 if (request_contains_null_bytes(csp, buf, len))
2030 /* NULL bytes found and dealt with, just hang up. */
2035 /* Does the request line look invalid? */
2036 if (client_protocol_is_unsupported(csp, req))
2039 * Yes. The request has already been
2040 * answered with a error response, the buffers
2041 * were freed and we're done with chatting.
2046 #ifdef FEATURE_FORCE_LOAD
2048 * If this request contains the FORCE_PREFIX and blocks
2049 * aren't enforced, get rid of it and set the force flag.
2051 if (strstr(req, FORCE_PREFIX))
2053 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2055 log_error(LOG_LEVEL_FORCE,
2056 "Ignored force prefix in request: \"%s\".", req);
2060 strclean(req, FORCE_PREFIX);
2061 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2062 csp->flags |= CSP_FLAG_FORCED;
2066 #endif /* def FEATURE_FORCE_LOAD */
2067 err = parse_http_request(req, http, csp);
2068 if (JB_ERR_OK != err)
2070 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2076 if (http->cmd == NULL)
2078 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2079 /* XXX: Use correct size */
2080 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2081 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2083 free_http_request(http);
2087 /* grab the rest of the client's headers */
2091 p = get_header(csp->iob);
2095 /* There are no additional headers to read. */
2102 * We didn't receive a complete header
2103 * line yet, get the rest of it.
2105 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2108 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2109 destroy_list(headers);
2113 if (add_to_iob(csp, buf, len))
2116 * If there is no memory left for buffering the
2117 * request, there is nothing we can do but hang up
2119 destroy_list(headers);
2126 * We were able to read a complete
2127 * header and can finaly enlist it.
2134 if (http->host == NULL)
2137 * If we still don't know the request destination,
2138 * the request is invalid or the client uses
2139 * Privoxy without its knowledge.
2141 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2144 * Our attempts to get the request destination
2145 * elsewhere failed or Privoxy is configured
2146 * to only accept proxy requests.
2148 * An error response has already been send
2149 * and we're done here.
2156 * Determine the actions for this URL
2158 #ifdef FEATURE_TOGGLE
2159 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2161 /* Most compatible set of actions (i.e. none) */
2162 init_current_action(csp->action);
2165 #endif /* ndef FEATURE_TOGGLE */
2167 get_url_actions(csp, http);
2171 * Save a copy of the original request for logging
2173 http->ocmd = strdup(http->cmd);
2175 if (http->ocmd == NULL)
2177 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2180 enlist(csp->headers, http->cmd);
2182 /* Append the previously read headers */
2183 list_append_list_unique(csp->headers, headers);
2184 destroy_list(headers);
2186 err = sed(csp, FILTER_CLIENT_HEADERS);
2187 if (JB_ERR_OK != err)
2189 assert(err == JB_ERR_PARSE);
2190 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2192 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2194 /* Check request line for rewrites. */
2195 if ((NULL == csp->headers->first->str)
2196 || (strcmp(http->cmd, csp->headers->first->str) &&
2197 (JB_ERR_OK != change_request_destination(csp))))
2200 * A header filter broke the request line - bail out.
2202 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2203 /* XXX: Use correct size */
2204 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2205 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2207 free_http_request(http);
2211 /* decide how to route the HTTP request */
2212 fwd = forward_url(csp, http);
2215 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2216 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2220 * build the http request to send to the server
2221 * we have to do one of the following:
2223 * create = use the original HTTP request to create a new
2224 * HTTP request that has either the path component
2225 * without the http://domainspec (w/path) or the
2226 * full orininal URL (w/url)
2227 * Note that the path and/or the HTTP version may
2228 * have been altered by now.
2230 * connect = Open a socket to the host:port of the server
2231 * and short-circuit server and client socket.
2233 * pass = Pass the request unchanged if forwarding a CONNECT
2234 * request to a parent proxy. Note that we'll be sending
2235 * the CFAIL message ourselves if connecting to the parent
2236 * fails, but we won't send a CSUCCEED message if it works,
2237 * since that would result in a double message (ours and the
2238 * parent's). After sending the request to the parent, we simply
2241 * here's the matrix:
2244 * +--------+--------+
2246 * 0 | create | connect|
2248 * Forwarding +--------+--------+
2250 * 1 | create | pass |
2252 * +--------+--------+
2256 if (http->ssl && connect_port_is_forbidden(csp))
2258 const char *acceptable_connect_ports =
2259 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2260 assert(NULL != acceptable_connect_ports);
2261 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2262 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2263 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2264 csp->action->flags |= ACTION_BLOCK;
2270 freez(csp->headers->first->str);
2271 build_request_line(csp, fwd, &csp->headers->first->str);
2274 hdr = list_to_text(csp->headers);
2277 /* FIXME Should handle error properly */
2278 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2282 * We have a request. Check if one of the crunchers wants it.
2284 if (crunch_response_triggered(csp, crunchers_all))
2287 * Yes. The client got the crunch response
2288 * and we are done here after cleaning up.
2291 list_remove_all(csp->headers);
2297 * The headers can't be removed earlier because
2298 * they were still needed for the referrer check
2299 * in case of CGI crunches.
2301 * XXX: Would it be worth to move the referrer check
2302 * into client_referrer() and set a flag if it's trusted?
2304 list_remove_all(csp->headers);
2306 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2308 if (fwd->forward_host)
2310 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2311 fwd->forward_host, fwd->forward_port, http->hostport);
2315 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2318 /* here we connect to the server, gateway, or the forwarder */
2320 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2321 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2323 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2324 forwarded_connect_retries, http->hostport);
2327 if (csp->sfd == JB_INVALID_SOCKET)
2329 if (fwd->type != SOCKS_NONE)
2332 rsp = error_response(csp, "forwarding-failed", errno);
2334 else if (errno == EINVAL)
2336 rsp = error_response(csp, "no-such-domain", errno);
2340 rsp = error_response(csp, "connect-failed", errno);
2341 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2346 /* Write the answer to the client */
2349 send_crunch_response(csp, rsp);
2356 if (fwd->forward_host || (http->ssl == 0))
2358 /* write the client's (modified) header to the server
2359 * (along with anything else that may be in the buffer)
2362 if (write_socket(csp->sfd, hdr, strlen(hdr))
2363 || (flush_socket(csp->sfd, csp->iob) < 0))
2365 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2368 rsp = error_response(csp, "connect-failed", errno);
2372 send_crunch_response(csp, rsp);
2382 * We're running an SSL tunnel and we're not forwarding,
2383 * so just send the "connect succeeded" message to the
2384 * client, flush the rest, and get out of the way.
2386 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2394 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2396 /* we're finished with the client's header */
2399 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2401 /* pass data between the client and server
2402 * until one or the other shuts down the connection.
2411 * FD_ZERO here seems to point to an errant macro which crashes.
2412 * So do this by hand for now...
2414 memset(&rfds,0x00,sizeof(fd_set));
2418 FD_SET(csp->cfd, &rfds);
2419 FD_SET(csp->sfd, &rfds);
2421 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2425 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2429 /* this is the body of the browser's request
2430 * just read it and write it.
2433 if (FD_ISSET(csp->cfd, &rfds))
2435 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2439 break; /* "game over, man" */
2442 if (write_socket(csp->sfd, buf, (size_t)len))
2444 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2451 * The server wants to talk. It could be the header or the body.
2452 * If `hdr' is null, then it's the header otherwise it's the body.
2453 * FIXME: Does `hdr' really mean `host'? No.
2457 if (FD_ISSET(csp->sfd, &rfds))
2460 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2464 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2466 if (http->ssl && (fwd->forward_host == NULL))
2469 * Just hang up. We already confirmed the client's CONNECT
2470 * request with status code 200 and unencrypted content is
2471 * no longer welcome.
2473 log_error(LOG_LEVEL_ERROR,
2474 "CONNECT already confirmed. Unable to tell the client about the problem.");
2477 else if (byte_count)
2480 * Just hang up. We already transmitted the original headers
2481 * and parts of the original content and therefore missed the
2482 * chance to send an error message (without risking data corruption).
2484 * XXX: we could retry with a fancy range request here.
2486 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2487 "Unable to tell the client about the problem.");
2491 rsp = error_response(csp, "connect-failed", errno);
2495 send_crunch_response(csp, rsp);
2501 /* Add a trailing zero. This lets filter_popups
2502 * use string operations.
2506 /* Normally, this would indicate that we've read
2507 * as much as the server has sent us and we can
2508 * close the client connection. However, Microsoft
2509 * in its wisdom has released IIS/5 with a bug that
2510 * prevents it from sending the trailing \r\n in
2511 * a 302 redirect header (and possibly other headers).
2512 * To work around this if we've haven't parsed
2513 * a full header we'll append a trailing \r\n
2514 * and see if this now generates a valid one.
2516 * This hack shouldn't have any impacts. If we've
2517 * already transmitted the header or if this is a
2518 * SSL connection, then we won't bother with this
2519 * hack. So we only work on partially received
2520 * headers. If we append a \r\n and this still
2521 * doesn't generate a valid header, then we won't
2522 * transmit anything to the client.
2527 if (server_body || http->ssl)
2530 * If we have been buffering up the document,
2531 * now is the time to apply content modification
2532 * and send the result to the client.
2536 p = execute_content_filter(csp, content_filter);
2538 * If the content filter fails, use the original
2539 * buffer and length.
2540 * (see p != NULL ? p : csp->iob->cur below)
2544 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2547 if (JB_ERR_OK != update_server_headers(csp))
2549 log_error(LOG_LEVEL_FATAL,
2550 "Failed to update server headers. after filtering.");
2553 hdr = list_to_text(csp->headers);
2556 /* FIXME Should handle error properly */
2557 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2560 if (write_socket(csp->cfd, hdr, strlen(hdr))
2561 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2563 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2573 break; /* "game over, man" */
2577 * This is NOT the body, so
2578 * Let's pretend the server just sent us a blank line.
2580 snprintf(buf, sizeof(buf), "\r\n");
2581 len = (int)strlen(buf);
2584 * Now, let the normal header parsing algorithm below do its
2585 * job. If it fails, we'll exit instead of continuing.
2592 * If this is an SSL connection or we're in the body
2593 * of the server document, just write it to the client,
2594 * unless we need to buffer the body for later content-filtering
2597 if (server_body || http->ssl)
2602 * If there is no memory left for buffering the content, or the buffer limit
2603 * has been reached, switch to non-filtering mode, i.e. make & write the
2604 * header, flush the iob and buf, and get out of the way.
2606 if (add_to_iob(csp, buf, len))
2611 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2613 hdr = list_to_text(csp->headers);
2617 * Memory is too tight to even generate the header.
2618 * Send our static "Out-of-memory" page.
2620 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2621 rsp = cgi_error_memory();
2622 send_crunch_response(csp, rsp);
2626 hdrlen = strlen(hdr);
2628 if (write_socket(csp->cfd, hdr, hdrlen)
2629 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2630 || (write_socket(csp->cfd, buf, (size_t)len)))
2632 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2639 * Reset the byte_count to the amount of bytes
2640 * we just flushed. len will be added a few lines below,
2641 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2643 byte_count = (size_t)flushed;
2645 content_filter = NULL;
2651 if (write_socket(csp->cfd, buf, (size_t)len))
2653 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2657 byte_count += (size_t)len;
2662 /* we're still looking for the end of the
2663 * server's header ... (does that make header
2664 * parsing an "out of body experience" ?
2668 * buffer up the data we just read. If that fails,
2669 * there's little we can do but send our static
2670 * out-of-memory page.
2672 if (add_to_iob(csp, buf, len))
2674 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2675 rsp = cgi_error_memory();
2676 send_crunch_response(csp, rsp);
2681 /* Convert iob into something sed() can digest */
2682 if (JB_ERR_PARSE == get_server_headers(csp))
2686 /* Well, we tried our MS IIS/5
2687 * hack and it didn't work.
2688 * The header is incomplete
2689 * and there isn't anything
2690 * we can do about it.
2692 log_error(LOG_LEVEL_INFO,
2693 "MS IIS5 hack didn't produce valid headers.");
2698 /* Since we have to wait for
2699 * more from the server before
2700 * we can parse the headers
2701 * we just continue here.
2707 /* Did we actually get anything? */
2708 if (NULL == csp->headers->first)
2710 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2711 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2712 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2713 free_http_request(http);
2717 assert(csp->headers->first->str);
2719 if (strncmpic(csp->headers->first->str, "HTTP", 4))
2722 * It doesn't look like a HTTP response:
2723 * tell the client and log the problem.
2725 if (strlen(csp->headers->first->str) > 30)
2727 csp->headers->first->str[30] = '\0';
2729 log_error(LOG_LEVEL_ERROR,
2730 "Invalid server or forwarder response. Starts with: %s",
2731 csp->headers->first->str);
2732 log_error(LOG_LEVEL_CLF,
2733 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2734 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2735 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2736 free_http_request(http);
2741 * We have now received the entire server header,
2742 * filter it and send the result to the client
2744 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2746 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2748 hdr = list_to_text(csp->headers);
2751 /* FIXME Should handle error properly */
2752 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2755 if (crunch_response_triggered(csp, crunchers_light))
2758 * One of the tags created by a server-header
2759 * tagger triggered a crunch. We already
2760 * delivered the crunch response to the client
2761 * and are done here after cleaning up.
2766 /* Buffer and pcrs filter this if appropriate. */
2768 if (!http->ssl) /* We talk plaintext */
2770 content_filter = get_filter_function(csp);
2773 * Only write if we're not buffering for content modification
2775 if (!content_filter)
2777 /* write the server's (modified) header to
2778 * the client (along with anything else that
2779 * may be in the buffer)
2782 if (write_socket(csp->cfd, hdr, strlen(hdr))
2783 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2785 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2787 /* the write failed, so don't bother
2788 * mentioning it to the client...
2789 * it probably can't hear us anyway.
2795 byte_count += (size_t)len;
2798 /* we're finished with the server's header */
2803 /* If this was a MS IIS/5 hack then it means
2804 * the server has already closed the
2805 * connection. Nothing more to read. Time
2810 log_error(LOG_LEVEL_INFO,
2811 "Closed server connection detected with MS IIS5 hack enabled.");
2818 return; /* huh? we should never get here */
2821 if (csp->content_length == 0)
2824 * If Privoxy didn't recalculate the
2825 * Content-Lenght, byte_count is still
2828 csp->content_length = byte_count;
2831 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2832 csp->ip_addr_str, http->ocmd, csp->content_length);
2836 /*********************************************************************
2840 * Description : This is little more than chat. We only "serve" to
2841 * to close any socket that chat may have opened.
2844 * 1 : csp = Current client state (buffers, headers, etc...)
2848 *********************************************************************/
2850 void serve(struct client_state *csp)
2851 #else /* ifndef AMIGA */
2852 static void serve(struct client_state *csp)
2853 #endif /* def AMIGA */
2856 close_socket(csp->cfd);
2858 if (csp->sfd != JB_INVALID_SOCKET)
2860 close_socket(csp->sfd);
2863 csp->flags &= ~CSP_FLAG_ACTIVE;
2869 /*********************************************************************
2871 * Function : server_thread
2873 * Description : We only exist to call `serve' in a threaded environment.
2876 * 1 : data = Current client state (buffers, headers, etc...)
2878 * Returns : Always 0.
2880 *********************************************************************/
2881 static int32 server_thread(void *data)
2883 serve((struct client_state *) data);
2890 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2891 /*********************************************************************
2895 * Description : Print usage info & exit.
2897 * Parameters : Pointer to argv[0] for identifying ourselves
2901 *********************************************************************/
2902 static void usage(const char *myname)
2904 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2908 #endif /* defined(unix) */
2911 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2912 #endif /* defined(unix) */
2913 "[--version] [configfile]\n"
2914 "Aborting\n", myname);
2919 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2922 /*********************************************************************
2924 * Function : initialize_mutexes
2926 * Description : Prepares mutexes if mutex support is available.
2930 * Returns : Void, exits in case of errors.
2932 *********************************************************************/
2933 static void initialize_mutexes(void)
2937 #ifdef FEATURE_PTHREAD
2939 * Prepare global mutex semaphores
2941 err = pthread_mutex_init(&log_mutex, 0);
2943 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2946 * XXX: The assumptions below are a bit naive
2947 * and can cause locks that aren't necessary.
2949 * For example older FreeBSD versions (< 6.x?)
2950 * have no gethostbyname_r, but gethostbyname is
2953 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2954 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2955 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2957 * XXX: should we use a single mutex for
2958 * localtime() and gmtime() as well?
2960 #ifndef HAVE_GMTIME_R
2961 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2962 #endif /* ndef HAVE_GMTIME_R */
2964 #ifndef HAVE_LOCALTIME_R
2965 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2966 #endif /* ndef HAVE_GMTIME_R */
2969 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2970 #endif /* ndef HAVE_RANDOM */
2971 #endif /* FEATURE_PTHREAD */
2974 * TODO: mutex support for mingw32 would be swell.
2979 printf("Fatal error. Mutex initialization failed: %s.\n",
2988 /*********************************************************************
2992 * Description : Load the config file and start the listen loop.
2993 * This function is a lot more *sane* with the `load_config'
2994 * and `listen_loop' functions; although it stills does
2995 * a *little* too much for my taste.
2998 * 1 : argc = Number of parameters (including $0).
2999 * 2 : argv = Array of (char *)'s to the parameters.
3001 * Returns : 1 if : can't open config file, unrecognized directive,
3002 * stats requested in multi-thread mode, can't open the
3003 * log file, can't open the jar file, listen port is invalid,
3004 * any load fails, and can't bind port.
3006 * Else main never returns, the process must be signaled
3007 * to terminate execution. Or, on Windows, use the
3008 * "File", "Exit" menu option.
3010 *********************************************************************/
3012 int real_main(int argc, const char *argv[])
3014 int main(int argc, const char *argv[])
3018 unsigned int random_seed;
3020 struct passwd *pw = NULL;
3021 struct group *grp = NULL;
3024 char *pre_chroot_nslookup_to_load_resolver = NULL;
3031 #if !defined(_WIN32)
3039 * Parse the command line arguments
3041 * XXX: simply printing usage information in case of
3042 * invalid arguments isn't particularly user friendly.
3044 while (++argc_pos < argc)
3047 /* Check to see if the service must be installed or uninstalled */
3048 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3050 const char *pName = argv[argc_pos] + 9;
3053 exit( (install_service(pName)) ? 0 : 1 );
3055 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3057 const char *pName = argv[argc_pos] + 11;
3060 exit((uninstall_service(pName)) ? 0 : 1);
3062 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3064 bRunAsService = TRUE;
3065 w32_set_service_cwd();
3066 atexit(w32_service_exit_notify);
3069 #endif /* defined(_WIN32) */
3072 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3074 if (strcmp(argv[argc_pos], "--help") == 0)
3079 else if(strcmp(argv[argc_pos], "--version") == 0)
3081 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3087 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3092 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3094 if (++argc_pos == argc) usage(argv[0]);
3095 pidfile = strdup(argv[argc_pos]);
3098 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3100 if (++argc_pos == argc) usage(argv[argc_pos]);
3102 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3105 if (NULL == (grp = getgrnam(p)))
3107 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3111 if (NULL == (pw = getpwnam(argv[argc_pos])))
3113 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3116 if (p != NULL) *--p = '\0';
3119 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3121 if (++argc_pos == argc) usage(argv[0]);
3122 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3125 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3129 #endif /* defined(unix) */
3131 else if (argc_pos + 1 != argc)
3134 * This is neither the last command line
3135 * option, nor was it recognized before,
3136 * therefore it must be invalid.
3142 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3144 configfile = argv[argc_pos];
3147 } /* -END- while (more arguments) */
3150 if ( *configfile != '/' )
3152 char cwd[BUFFER_SIZE];
3154 size_t abs_file_size;
3156 /* make config-filename absolute here */
3157 if (NULL == getcwd(cwd, sizeof(cwd)))
3159 perror("failed to get current working directory");
3164 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3165 basedir = strdup(cwd);
3167 if (NULL == basedir ||
3168 NULL == (abs_file = malloc(abs_file_size)))
3170 perror("malloc failed");
3173 strlcpy(abs_file, basedir, abs_file_size);
3174 strlcat(abs_file, "/", abs_file_size );
3175 strlcat(abs_file, configfile, abs_file_size);
3176 configfile = abs_file;
3178 #endif /* defined unix */
3182 clients->next = NULL;
3184 /* XXX: factor out initialising after the next stable release. */
3187 #elif defined(_WIN32)
3191 /* Prepare mutexes if supported and necessary. */
3192 initialize_mutexes();
3194 /* Enable logging until further notice. */
3195 init_log_module(Argv[0]);
3197 random_seed = (unsigned int)time(NULL);
3199 srandom(random_seed);
3200 #elif defined (_WIN32)
3202 * See pick_from_range() in miscutil.c for details.
3204 log_error(LOG_LEVEL_INFO,
3205 "No thread-safe PRNG implemented for your platform. "
3206 "Using weak \'randomization\' factor which will "
3207 "limit the already questionable usefulness of "
3208 "header-time-randomizing actions (disabled by default).");
3211 #endif /* ifdef HAVE_RANDOM */
3214 * Unix signal handling
3216 * Catch the abort, interrupt and terminate signals for a graceful exit
3217 * Catch the hangup signal so the errlog can be reopened.
3218 * Ignore the broken pipe signals (FIXME: Why?)
3220 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3223 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3224 const int ignored_signals[] = { SIGPIPE, 0 };
3226 for (idx = 0; catched_signals[idx] != 0; idx++)
3228 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3229 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3231 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3232 #endif /* ifdef sun */
3234 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3238 for (idx = 0; ignored_signals[idx] != 0; idx++)
3240 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3242 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3247 #else /* ifdef _WIN32 */
3248 # ifdef _WIN_CONSOLE
3250 * We *are* in a windows console app.
3251 * Print a verbose messages about FAQ's and such
3253 printf("%s", win32_blurb);
3254 # endif /* def _WIN_CONSOLE */
3255 #endif /* def _WIN32 */
3258 /* Initialize the CGI subsystem */
3259 cgi_init_error_messages();
3262 * If runnig on unix and without the --nodaemon
3263 * option, become a daemon. I.e. fork, detach
3264 * from tty and get process group leadership
3277 if ( pid < 0 ) /* error */
3282 else if ( pid != 0 ) /* parent */
3287 * must check for errors
3288 * child died due to missing files aso
3291 wpid = waitpid( pid, &status, WNOHANG );
3300 /* Should be more portable, but not as well tested */
3305 #else /* ndef __FreeBSD__ */
3307 #endif /* ndef __FreeBSD__ */
3308 fd = open("/dev/tty", O_RDONLY);
3311 /* no error check here */
3312 ioctl( fd, TIOCNOTTY,0 );
3317 * stderr (fd 2) will be closed later on, when the
3318 * log file has been parsed.
3325 } /* -END- if (!no_daemon) */
3328 * As soon as we have written the PID file, we can switch
3329 * to the user and group ID indicated by the --user option
3335 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3337 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3343 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3345 /* Read the time zone file from /etc before doing chroot. */
3347 if (NULL != pre_chroot_nslookup_to_load_resolver
3348 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3350 /* Initialize resolver library. */
3351 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3353 if (chroot(pw->pw_dir) < 0)
3355 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3359 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3362 if (setuid(pw->pw_uid))
3364 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3368 char putenv_dummy[64];
3370 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3371 if (putenv(putenv_dummy) != 0)
3373 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3376 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3377 if (putenv(putenv_dummy) != 0)
3379 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3385 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3388 #endif /* defined unix */
3391 /* This will be FALSE unless the command line specified --service
3395 /* Yup, so now we must attempt to establish a connection
3396 * with the service dispatcher. This will only work if this
3397 * process was launched by the service control manager to
3398 * actually run as a service. If this isn't the case, i've
3399 * known it take around 30 seconds or so for the call to return.
3402 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3403 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3405 /* Service has run, and at this point is now being stopped, so just return */
3410 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3412 /* An error occurred. Usually it's because --service was wrongly specified
3413 * and we were unable to connect to the Service Control Dispatcher because
3414 * it wasn't expecting us and is therefore not listening.
3416 * For now, just continue below to call the listen_loop function.
3419 #endif /* def _WIN32 */
3429 /*********************************************************************
3431 * Function : bind_port_helper
3433 * Description : Bind the listen port. Handles logging, and aborts
3437 * 1 : config = Privoxy configuration. Specifies port
3440 * Returns : Port that was opened.
3442 *********************************************************************/
3443 static jb_socket bind_port_helper(struct configuration_spec * config)
3448 if (config->haddr == NULL)
3450 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3455 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3456 config->hport, config->haddr);
3459 result = bind_port(config->haddr, config->hport, &bfd);
3466 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3467 "There may be another Privoxy or some other "
3468 "proxy running on port %d",
3469 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3470 config->hport, config->hport);
3473 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3474 "The hostname is not resolvable",
3475 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3478 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3479 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3482 /* shouldn't get here */
3483 return JB_INVALID_SOCKET;
3486 config->need_bind = 0;
3493 /* Without this simple workaround we get this compiler warning from _beginthread
3494 * warning C4028: formal parameter 1 different from declaration
3496 void w32_service_listen_loop(void *p)
3500 #endif /* def _WIN32 */
3503 /*********************************************************************
3505 * Function : listen_loop
3507 * Description : bind the listen port and enter a "FOREVER" listening loop.
3513 *********************************************************************/
3514 static void listen_loop(void)
3516 struct client_state *csp = NULL;
3518 struct configuration_spec * config;
3520 config = load_config();
3522 bfd = bind_port_helper(config);
3524 #ifdef FEATURE_GRACEFUL_TERMINATION
3525 while (!g_terminate)
3530 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3531 while (waitpid(-1, NULL, WNOHANG) > 0)
3533 /* zombie children */
3535 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3538 * Free data that was used by died threads
3544 * Re-open the errlog after HUP signal
3546 if (received_hup_signal)
3548 init_error_log(Argv[0], config->logfile);
3549 received_hup_signal = 0;
3554 #ifdef FEATURE_COOKIE_JAR
3556 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3557 * already open. I think unload_configfile() is not being run, which should do
3558 * this work. Until that can get resolved, we'll use this workaround.
3562 if (csp->config->jar)
3564 fclose(csp->config->jar);
3565 csp->config->jar = NULL;
3567 #endif /* FEATURE_COOKIE_JAR */
3568 #endif /* __OS2__ */
3570 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3572 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3576 csp->flags |= CSP_FLAG_ACTIVE;
3577 csp->sfd = JB_INVALID_SOCKET;
3579 csp->config = config = load_config();
3581 if ( config->need_bind )
3584 * Since we were listening to the "old port", we will not see
3585 * a "listen" param change until the next IJB request. So, at
3586 * least 1 more request must be made for us to find the new
3587 * setting. I am simply closing the old socket and binding the
3590 * Which-ever is correct, we will serve 1 more page via the
3591 * old settings. This should probably be a "show-proxy-args"
3592 * request. This should not be a so common of an operation
3593 * that this will hurt people's feelings.
3598 bfd = bind_port_helper(config);
3601 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3603 if (!accept_connection(csp, bfd))
3605 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3618 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3621 #ifdef FEATURE_TOGGLE
3622 if (global_toggle_state)
3623 #endif /* def FEATURE_TOGGLE */
3625 csp->flags |= CSP_FLAG_TOGGLED_ON;
3628 if (run_loader(csp))
3630 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3631 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3635 if (block_acl(NULL,csp))
3637 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3638 close_socket(csp->cfd);
3642 #endif /* def FEATURE_ACL */
3644 /* add it to the list of clients */
3645 csp->next = clients->next;
3646 clients->next = csp;
3648 if (config->multi_threaded)
3652 /* this is a switch () statment in the C preprocessor - ugh */
3653 #undef SELECTED_ONE_OPTION
3655 /* Use Pthreads in preference to native code */
3656 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3657 #define SELECTED_ONE_OPTION
3659 pthread_t the_thread;
3660 pthread_attr_t attrs;
3662 pthread_attr_init(&attrs);
3663 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3664 errno = pthread_create(&the_thread, &attrs,
3665 (void * (*)(void *))serve, csp);
3666 child_id = errno ? -1 : 0;
3667 pthread_attr_destroy(&attrs);
3671 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3672 #define SELECTED_ONE_OPTION
3673 child_id = _beginthread(
3674 (void (*)(void *))serve,
3679 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3680 #define SELECTED_ONE_OPTION
3681 child_id = _beginthread(
3682 (void(* _Optlink)(void*))serve,
3688 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3689 #define SELECTED_ONE_OPTION
3691 thread_id tid = spawn_thread
3692 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3694 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3696 child_id = (int) tid;
3705 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3706 #define SELECTED_ONE_OPTION
3707 csp->cfd = ReleaseSocket(csp->cfd, -1);
3710 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3711 NP_Output, Output(),
3712 NP_CloseOutput, FALSE,
3713 NP_Name, (ULONG)"privoxy child",
3717 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3718 NP_Output, Output(),
3719 NP_CloseOutput, FALSE,
3720 NP_Name, (ULONG)"privoxy child",
3721 NP_StackSize, 200*1024,
3727 ((struct Task *)child_id)->tc_UserData = csp;
3728 Signal((struct Task *)child_id, SIGF_SINGLE);
3733 #if !defined(SELECTED_ONE_OPTION)
3736 /* This block is only needed when using fork().
3737 * When using threads, the server thread was
3738 * created and run by the call to _beginthread().
3740 if (child_id == 0) /* child */
3743 #ifdef FEATURE_TOGGLE
3744 int inherited_toggle_state = global_toggle_state;
3745 #endif /* def FEATURE_TOGGLE */
3750 * If we've been toggled or we've blocked the request, tell Mom
3753 #ifdef FEATURE_TOGGLE
3754 if (inherited_toggle_state != global_toggle_state)
3756 rc |= RC_FLAG_TOGGLED;
3758 #endif /* def FEATURE_TOGGLE */
3760 #ifdef FEATURE_STATISTICS
3761 if (csp->flags & CSP_FLAG_REJECTED)
3763 rc |= RC_FLAG_BLOCKED;
3765 #endif /* ndef FEATURE_STATISTICS */
3769 else if (child_id > 0) /* parent */
3771 /* in a fork()'d environment, the parent's
3772 * copy of the client socket and the CSP
3776 #if !defined(_WIN32) && !defined(__CYGWIN__)
3778 wait( &child_status );
3781 * Evaluate child's return code: If the child has
3782 * - been toggled, toggle ourselves
3783 * - blocked its request, bump up the stats counter
3786 #ifdef FEATURE_TOGGLE
3787 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3789 global_toggle_state = !global_toggle_state;
3791 #endif /* def FEATURE_TOGGLE */
3793 #ifdef FEATURE_STATISTICS
3795 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3799 #endif /* def FEATURE_STATISTICS */
3801 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3802 close_socket(csp->cfd);
3803 csp->flags &= ~CSP_FLAG_ACTIVE;
3807 #undef SELECTED_ONE_OPTION
3808 /* end of cpp switch () */
3810 if (child_id < 0) /* failed */
3812 char buf[BUFFER_SIZE];
3814 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3816 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3818 write_socket(csp->cfd, buf, strlen(buf));
3819 close_socket(csp->cfd);
3820 csp->flags &= ~CSP_FLAG_ACTIVE;
3831 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3833 /* Clean up. Aim: free all memory (no leaks) */
3834 #ifdef FEATURE_GRACEFUL_TERMINATION
3836 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3838 unload_current_config_file();
3839 unload_current_actions_file();
3840 unload_current_re_filterfile();
3841 #ifdef FEATURE_TRUST
3842 unload_current_trust_file();
3845 if (config->multi_threaded)
3852 } while ((clients->next != NULL) && (--i > 0));
3856 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3867 #ifdef FEATURE_COOKIE_JAR
3868 if (NULL != config->jar)
3870 fclose(config->jar);
3874 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3875 /* Cleanup - remove taskbar icon etc. */
3880 #endif /* FEATURE_GRACEFUL_TERMINATION */