1 const char jcc_rcs[] = "$Id: jcc.c,v 1.188 2008/10/09 18:21:41 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 1.188 2008/10/09 18:21:41 fabiankeil
37 * Flush work-in-progress changes to keep outgoing connections
38 * alive where possible. Incomplete and mostly #ifdef'd out.
40 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
41 * Add mutex lock support for _WIN32.
43 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
44 * Prepare for critical sections on Windows by adding a
45 * layer of indirection before the pthread mutex functions.
47 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
48 * Remove FEATURE_COOKIE_JAR.
50 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
51 * - Silence LLVM/Clang complaint.
52 * - Make received_hup_signal static.
53 * - Hide definitions for basedir, pidfile and received_hup_signal
54 * from __EMX__ as they only seem to be used in case of #ifdef unix.
56 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
57 * Accept Shoutcast responses again. Problem reported
58 * and fix suggested by Stefan in #2062860.
60 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
61 * Fix possible NULL-pointer dereference reported
62 * by din_a4 in #2003937. Pointy hat to me.
64 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
65 * Streamline sed()'s prototype and declare
66 * the header parse and add structures static.
68 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
69 * - Mark csp as immutable for send_crunch_response().
70 * - Fix comment spelling.
72 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
73 * Factor update_server_headers() out of sed(), ditch the
74 * first_run hack and make server_patterns_light static.
76 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
77 * Don't provide get_header() with the whole client state
78 * structure when it only needs access to csp->iob.
80 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
81 * Make the "read the rest of the headers" loop a bit more readable.
83 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
84 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
85 * - Remove useless comment.
87 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
88 * Fix comment grammar.
90 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
91 * Remove the pointless buffer in client_protocol_is_unsupported().
93 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
94 * Least-effort fix for bug #1821930 (reported by Lee):
95 * If the response doesn't look like HTTP,
96 * tell the client and log the problem.
98 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
99 * Don't pass the whole csp structure to flush_socket()
100 * when it only needs a file descriptor and a buffer.
102 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
103 * Remove kill-popups action.
105 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
106 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
108 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
109 * Remove the treat-forbidden-connects-like-blocks action. We now
110 * use the "blocked" page for forbidden CONNECT requests by default.
112 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
113 * Also use shiny new connect_port_is_forbidden() in jcc.c.
115 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
116 * Rename url_actions() to get_url_actions() and let it
117 * use the standard parameter ordering.
119 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
120 * Let forward_url() use the standard parameter ordering
121 * and mark its second parameter immutable.
123 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
124 * Remove the "Listening ... for local connections only" log message.
125 * Whether or not remote connections are able to reach Privoxy is up
126 * to the operating system.
128 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
129 * Prevent the log messages for CONNECT requests to unacceptable
130 * ports from printing the limit-connect argument as [null] if
131 * limit-connect hasn't been explicitly enabled.
133 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
134 * Make sure all console-mode apps get a usage() instance
136 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
137 * Reword NO_SERVER_DATA_RESPONSE to make it harder
138 * to misunderstand what the message is all about.
140 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
141 * Unbreak trustfile which previously didn't work without
142 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
144 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
145 * Plug memory leak. Spotted by Valgrind, triggered by
146 * Privoxy-Regression-Test feeding proxyfuzz.py.
148 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
149 * In the HTTP snipplets, refer to the client as client.
151 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
152 * Emit a log message when activating the MS IIS5 hack.
154 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
155 * Log the "weak randomization factor" warning only
156 * once for mingw32 and provide some more details.
158 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
159 * Initialize log module after initializing mutexes, future
160 * deadlocks in that code should now work cross-platform.
162 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
163 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
164 * Reported by Bert van Leeuwen in BR#1818808.
166 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
167 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
169 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
170 * When in daemon mode, close stderr after the configuration file has been
171 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
173 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
174 * - Fix a crash when parsing invalid requests whose first header
175 * is rejected by get_header(). Regression (re?)introduced
176 * in r1.143 by yours truly.
177 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
178 * client_cookie_adder() to make sure send-vanilla-wafer can be
179 * controlled through tags (and thus regression-tested).
181 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
182 * - Move get_filter_function() from jcc.c to filters.c
183 * so the filter functions can be static.
184 * - Don't bother filtering body-less responses.
186 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
187 * Execute content filters through execute_content_filter().
189 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
190 * Initialize req to NULL to make sure it's defined if the
191 * first read_socket() call fails. Reported by icmp30.
193 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
194 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
195 * extensive comments moved to user manual.
197 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
198 * Don't crash if a broken header filter wiped out the request line.
200 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
201 * Fix byte_count calculation in case of flushes
202 * and don't parse the server headers a second time.
204 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
205 * - If there's a connection problem after we already forwarded
206 * parts of the original content, just hang up. Fixes BR#1776724.
207 * - Fix warnings about unused code on mingw32.
208 * - In case of flushes, calculate the byte count
209 * less incorrectly (I think).
211 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
212 * Add some more prototypes for static functions.
214 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
215 * Comment out request_contains_null_bytes() until it's used again.
217 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
218 * #1763173 from Stefan Huehner: s@const static@static const@
219 * and declare some more functions static.
221 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
222 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
223 * and explain why they were rejected in the first place.
224 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
225 * CONNECT requests. The request line was missing.
226 * - Add two more XXX reminders as we don't have enough already.
228 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
229 * As Hal noticed, checking dispatch_cgi() as the last cruncher
230 * looks like a bug if CGI requests are blocked unintentionally,
231 * so don't do it unless the user enabled the new config option
232 * "allow-cgi-request-crunching".
234 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
235 * - Allow to rewrite the request destination behind the client's back.
236 * - Turn the weird-looking unconditional for loop that
237 * reads the client request into a conditional while loop.
238 * Move the stuff that only runs once out of the loop.
239 * - Move parts of chat(), server_content_type() and the
240 * necessary stuff to fix BR#1750917 into get_filter_function().
242 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
243 * Temporary workaround for BR#1730105.
245 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
246 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
247 * deadlocks and crashes on OpenBSD and possibly other OS with neither
248 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
249 * Thanks to Ralf Horstmann for report and solution.
251 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
252 * Add forward-override{} to change the forwarding settings through
253 * action sections. This is mainly interesting to forward different
254 * clients differently (for example based on User-Agent or request
257 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
258 * - Let usage() mention the --chroot parameter.
259 * - Use read_socket() consistently and always leave
260 * the last buffer byte alone, even in cases where
261 * null termination (currently) doesn't matter.
263 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
264 * - Fix config file loading on Unix if no config file is specified.
265 * Since r1.97 Privoxy would always interpret the last argument as
266 * config file, even if it's a valid command line option.
267 * - Abort in case of unrecognized command line options. Closes #1719696.
268 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
269 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
271 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
272 * - Don't rerun crunchers that only depend on the request URL.
273 * - Don't count redirects and CGI requests as "blocked requests".
275 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
276 * Support crunching based on tags created by server-header taggers.
278 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
279 * Make HTTP snippets static (again). Add a Content-Type for those
280 * with content so the browser doesn't guess it based on the URL.
282 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
283 * Move crunching and request line rebuilding out of chat().
285 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
286 * Introduce tags as alternative way to specify which
287 * actions apply to a request. At the moment tags can be
288 * created based on client and server headers.
290 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
291 * Don't CLF-log CONNECT requests twice.
293 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
294 * Log the source address for ACL-related connection drops.
296 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
297 * New config option: enforce-blocks.
299 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
300 * - Move null byte check into separate function.
301 * - Don't confuse the client with error pages
302 * if a CONNECT request was already confirmed.
304 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
305 * Speed up NULL byte escaping and only log the complete
306 * NULL byte requests with header debugging enabled.
308 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
309 * Answer requests that contain NULL bytes with
310 * a custom response instead of waiting for more
311 * data until the client eventually hangs up.
313 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
314 * - Move delivery and logging of crunched responses
315 * from chat() into send_crunch_response().
316 * - Display the reason for generating http_responses.
317 * - Log the content length for LOG_LEVEL_CLF correctly
318 * (still incorrect for some fixed responses).
319 * - Reword an incorrect comment about
320 * treat-forbidden-connects-like-blocks violating
322 * - Add some log messages.
324 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
325 * Move mutex initialization into separate
326 * function and exit in case of errors.
328 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
329 * - Start to reduce chat()'s line count and move
330 * parts of it into separate functions.
331 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
333 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
334 * - Add Proxy-Agent header to HTTP snippets that are
335 * supposed to reach HTTP clients only.
336 * - Made a few CONNECT log messages more descriptive.
337 * - Catch completely empty server responses (as seen
338 * with Tor's fake ".noconnect" top level domain).
339 * - Use shiny new "forwarding-failed" template for socks errors.
341 * Revision 1.118 2007/01/07 07:43:43 joergs
342 * AmigaOS4 support added.
344 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
345 * Added config option accept-intercepted-requests
346 * and disabled it by default.
348 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
349 * Reverted parts of my last commit
350 * to keep error handling working.
352 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
353 * Fixed gcc43 conversion warnings.
355 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
356 * Fix -pedantic ISO C warning about converting
357 * from function pointer to object pointer.
359 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
360 * Silence compiler warning I introduced with my last commit.
362 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
363 * Mutex protect rand() if POSIX threading
364 * is used, warn the user if that's not possible
365 * and stop using it on _WIN32 where it could
368 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
369 * Don't prevent core dumps by catching SIGABRT.
370 * It's rude and makes debugging unreasonable painful.
372 * Revision 1.110 2006/12/13 14:52:53 etresoft
373 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
375 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
376 * Privoxy is now able to run as intercepting
377 * proxy in combination with any packet filter
378 * that does the port redirection. The destination
379 * is extracted from the "Host:" header which
380 * should be available for nearly all requests.
382 * Moved HTTP snipplets into jcc.c.
383 * Added error message for gopher proxy requests.
385 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
386 * Only unlink the pidfile if it's actually used.
388 * Change order of interception checks to make
389 * it possible to block or redirect requests for
392 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
393 * Make pthread mutex locking more generic. Instead of
394 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
395 * and use mutex locking unless there is an _r function
396 * available. Better safe than sorry.
398 * Fixes "./configure --disable-pthread" and should result
399 * in less threading-related problems on pthread-using platforms,
400 * but it still doesn't fix BR#1122404.
402 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
403 * Move pthread.h inclusion from jcc.c to jcc.h.
404 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
406 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
407 * Don't exit after receiving the second SIGHUP on Solaris.
409 * Fixes BR 1052235, but the same problem may exist on other
410 * systems. Once 3.0.6 is out we should use sigset()
411 * where available and see if it breaks anything.
413 * Revision 1.104 2006/09/23 13:26:38 roro
414 * Replace TABs by spaces in source code.
416 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
417 * Fix +redirect{}. Didn't work with -fast-redirects.
419 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
420 * Respond with 400 and a short text message
421 * if the client tries to use Privoxy as FTP proxy.
423 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
424 * Make number of retries in case of forwarded-connect problems
425 * a config file option (forwarded-connect-retries) and use 0 as
428 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
429 * Set random(3) seed.
431 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
432 * Follow the OpenBSD port's lead and protect the resolve
433 * functions on OpenBSD as well.
435 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
436 * --user fix. Only use the user as group if no group is specified.
437 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
439 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
440 * Windows service (re-)integration
444 * --install[:service_name]
445 * --uninstall[:service_name]
448 * They work as follows:
449 * --install will create a service for you and then terminate.
450 * By default the service name will be "privoxy" (without the quotes).
451 * However you can run multiple services if you wish, just by adding
452 * a colon and then a name (no spaces).
454 * --uninstall follows the exact same rules a --install.
456 * --service is used when the program is executed by the service
457 * control manager, and in normal circumstances would never be
458 * used as a command line argument.
460 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
461 * Windows service integration
463 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
464 * Incorporate Fabian Keil's patch work:
465 http://www.fabiankeil.de/sourcecode/privoxy/
467 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
468 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
469 * with what was really the latest development (the v_3_0_branch branch)
471 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
472 * Thanks to Jindrich Makovicka for a race condition fix for the log
473 * file. The race condition remains for non-pthread implementations.
474 * Reference patch #1175720.
476 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
477 * Add the ability to check jpeg images for invalid
478 * lengths of comment blocks. Defensive strategy
479 * against the exploit:
480 * Microsoft Security Bulletin MS04-028
481 * Buffer Overrun in JPEG Processing (GDI+) Could
482 * Allow Code Execution (833987)
483 * Enabled with +inspect-jpegs in actions files.
485 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
486 * - Fixed usage info for non-unix platforms
487 * - Fixed small cmdline parsing bug
489 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
490 * Diagnostics: Now preserve the returncode of pthread_create
491 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
493 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
494 * No longer ignore SIGCHLD. Fixes bug #769381
496 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
497 * Close jarfile on graceful exit, remove stray line
499 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
500 * Cosmetics: Killed a warning, a typo and an allocation left at exit
502 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
503 * No longer rely on non-POSIX.1 extensions of getcwd().
506 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
507 * Replaced setenv() by posix-compliant putenv()
508 * Thanks to Neil McCalden (nmcc AT users.sf.net).
510 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
511 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
513 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
514 * Clean-up and extension of improvements for forked mode:
515 * - Child's return code now consists of flags RC_FLAG_*
516 * - Reporting toggle to parent now properly #ifdef'ed
517 * - Children now report blocking to parent. This enables
518 * statistics in forked mode
520 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
521 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
522 * Children now report having been toggled through _exit(17), parents
523 * watch for that code and toggle themselves if found.
525 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
526 * Wrapping all *_r functions (the non-_r versions of them) with
527 * mutex semaphores for OSX. Hopefully this will take care of all
528 * of those pesky crash reports.
530 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
531 * Fixed two mostly harmless mem leaks
533 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
534 * Fix: Head of global clients list now initialized to NULL
536 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
537 * Added basic support for OPTIONS and TRACE HTTP methods:
538 * - New interceptor direct_response() added in chat().
539 * - sed() moved to earlier in the process, so that the
540 * Host: header is evaluated before actions and forwarding
543 * Revision 1.92 2002/05/08 16:00:46 oes
544 * Chat's buffer handling:
545 * - Fixed bug with unchecked out-of-mem conditions
546 * while reading client request & server headers
547 * - No longer predict if the buffer limit will be exceeded
548 * in the next read -- check add_to_iob's new
549 * return code. If buffer couldn't be extended
550 * (policy or out-of-mem) while
551 * - reading from client: abort
552 * - reading server headers: send error page
553 * - buffering server body for filter: flush,
554 * and if that fails: send error page
556 * Revision 1.91 2002/04/08 20:35:58 swa
559 * Revision 1.90 2002/04/02 14:57:28 oes
560 * Made sending wafers independent of FEATURE_COOKIE_JAR
562 * Revision 1.89 2002/03/31 17:18:59 jongfoster
563 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
565 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
566 * More compiler warning message maintenance
568 * Revision 1.87 2002/03/26 22:29:54 swa
569 * we have a new homepage!
571 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
572 * Workaround for closing the jarfile before load_config() comes around again
574 * Revision 1.85 2002/03/24 15:23:33 jongfoster
577 * Revision 1.84 2002/03/24 13:25:43 swa
578 * name change related issues
580 * Revision 1.83 2002/03/16 23:54:06 jongfoster
581 * Adding graceful termination feature, to help look for memory leaks.
582 * If you enable this (which, by design, has to be done by hand
583 * editing config.h) and then go to http://i.j.b/die, then the program
584 * will exit cleanly after the *next* request. It should free all the
585 * memory that was used.
587 * Revision 1.82 2002/03/13 00:27:05 jongfoster
590 * Revision 1.81 2002/03/12 01:42:50 oes
591 * Introduced modular filters
593 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
594 * OS/2 port maintenance:
595 * - Fixed EMX build - it had decayed a little
596 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
597 * substituted a memset for now.
599 * Revision 1.79 2002/03/09 20:03:52 jongfoster
600 * - Making various functions return int rather than size_t.
601 * (Undoing a recent change). Since size_t is unsigned on
602 * Windows, functions like read_socket that return -1 on
603 * error cannot return a size_t.
605 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
606 * crashes, and also frequently caused JB to jump to 100%
607 * CPU and stay there. (Because it thought it had just
608 * read ((unsigned)-1) == 4Gb of data...)
610 * - The signature of write_socket has changed, it now simply
611 * returns success=0/failure=nonzero.
613 * - Trying to get rid of a few warnings --with-debug on
614 * Windows, I've introduced a new type "jb_socket". This is
615 * used for the socket file descriptors. On Windows, this
616 * is SOCKET (a typedef for unsigned). Everywhere else, it's
617 * an int. The error value can't be -1 any more, so it's
618 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
619 * Windows it maps to the #define INVALID_SOCKET.)
621 * - The signature of bind_port has changed.
623 * Revision 1.78 2002/03/08 21:35:04 oes
624 * Added optional group supplement to --user option. Will now use default group of user if no group given
626 * Revision 1.77 2002/03/07 03:52:06 oes
627 * - Fixed compiler warnings etc
628 * - Improved handling of failed DNS lookups
630 * Revision 1.76 2002/03/06 22:54:35 jongfoster
631 * Automated function-comment nitpicking.
633 * Revision 1.75 2002/03/06 10:02:19 oes
634 * Fixed stupid bug when --user was not given
636 * Revision 1.74 2002/03/06 00:49:31 jongfoster
637 * Fixing warning on Windows
638 * Making #ifdefs that refer to the same variable consistently
639 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
641 * Revision 1.73 2002/03/05 23:57:30 hal9
642 * Stray character 's' on line 1618 was breaking build.
644 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
645 * - Re-enable OS/2 building after new parms were added
646 * - Fix false out of memory report when resolving CGI templates when no IP
647 * address is available of failed attempt (a la no such domain)
649 * Revision 1.71 2002/03/05 18:13:56 oes
650 * Added --user option
652 * Revision 1.70 2002/03/05 04:52:42 oes
653 * Deleted non-errlog debugging code
655 * Revision 1.69 2002/03/04 23:50:00 jongfoster
656 * Splitting off bind_port() call into bind_port_helper(), with
659 * Revision 1.68 2002/03/04 20:17:32 oes
662 * Revision 1.67 2002/03/04 18:18:57 oes
663 * - Removed _DEBUG mode
664 * - Cleand up cmdline parsing
665 * - Introduced --no-daemon, --pidfile options
666 * - Cleaned up signal handling:
667 * - Terminate cleanly on INT, TERM and ABRT
668 * - Schedule logfile for re-opening on HUP
669 * - Ignore CHLD and PIPE
670 * - Leave the rest with their default handlers
671 * - Uniform handler registration
672 * - Added usage() function
673 * - Played styleguide police
675 * Revision 1.66 2002/03/03 15:06:55 oes
676 * Re-enabled automatic config reloading
678 * Revision 1.65 2002/03/03 14:49:11 oes
679 * Fixed CLF logging: Now uses client's original HTTP request
681 * Revision 1.64 2002/03/03 09:18:03 joergs
682 * Made jumbjuster work on AmigaOS again.
684 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
685 * Clean up a little CRLF unpleasantness that suddenly appeared
687 * Revision 1.62 2002/02/20 23:17:23 jongfoster
688 * Detecting some out-of memory conditions and exiting with a log message.
690 * Revision 1.61 2002/01/17 21:01:52 jongfoster
691 * Moving all our URL and URL pattern parsing code to urlmatch.c.
693 * Revision 1.60 2001/12/30 14:07:32 steudten
694 * - Add signal handling (unix)
695 * - Add SIGHUP handler (unix)
696 * - Add creation of pidfile (unix)
697 * - Add action 'top' in rc file (RH)
698 * - Add entry 'SIGNALS' to manpage
699 * - Add exit message to logfile (unix)
701 * Revision 1.59 2001/12/13 14:07:18 oes
702 * Fixed Bug: 503 error page now sent OK
704 * Revision 1.58 2001/11/30 23:37:24 jongfoster
705 * Renaming the Win32 config file to config.txt - this is almost the
706 * same as the corresponding UNIX name "config"
708 * Revision 1.57 2001/11/16 00:47:43 jongfoster
709 * Changing the tty-disconnection code to use setsid().
711 * Revision 1.56 2001/11/13 20:20:54 jongfoster
712 * Tabs->spaces, fixing a bug with missing {} around an if()
714 * Revision 1.55 2001/11/13 20:14:53 jongfoster
715 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
717 * Revision 1.54 2001/11/07 00:03:14 steudten
718 * Give reliable return value if an error
719 * occurs not just 0 with new daemon mode.
721 * Revision 1.53 2001/11/05 21:41:43 steudten
722 * Add changes to be a real daemon just for unix os.
723 * (change cwd to /, detach from controlling tty, set
724 * process group and session leader to the own process.
726 * Add some fatal-error log message for failed malloc().
727 * Add '-d' if compiled with 'configure --with-debug' to
728 * enable debug output.
730 * Revision 1.52 2001/10/26 20:11:20 jongfoster
731 * Fixing type mismatch
733 * Revision 1.51 2001/10/26 17:38:28 oes
736 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
737 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
738 * threads to call select() simultaneously. So, it's time to do a real, live,
739 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
740 * (native). Both versions will work, but using __OS2__ offers multi-threading.
742 * Revision 1.49 2001/10/23 21:41:35 jongfoster
743 * Added call to initialize the (statically-allocated of course)
744 * "out of memory" CGI response.
746 * Revision 1.48 2001/10/10 19:56:46 jongfoster
747 * Moving some code that wasn't cookie-related out of an #ifdef
750 * Revision 1.47 2001/10/10 16:44:36 oes
751 * Added CONNECT destination port limitation check
753 * Revision 1.46 2001/10/08 15:17:41 oes
754 * Re-enabled SSL forwarding
756 * Revision 1.45 2001/10/07 15:42:11 oes
757 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
759 * Moved downgrading of the HTTP version from parse_http_request to
760 * chat(), since we can't decide if it is necessary before we have
761 * determined the actions for the URL. The HTTP command is now
762 * *always* re-built so the repairs need no longer be special-cased.
764 * filter_popups now gets a csp pointer so it can raise the new
765 * CSP_FLAG_MODIFIED flag.
769 * Added configurable size limit for the IOB. If the IOB grows so
770 * large that the next read would exceed the limit, the header
771 * is generated, and the header & unfiltered buffer are flushed
772 * to the client. Chat then continues in non-buffering,
773 * non-filtering body mode.
775 * Revision 1.44 2001/10/02 18:13:57 oes
778 * Revision 1.43 2001/10/02 15:32:13 oes
779 * Moved generation of hdr
781 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
782 * Cleaning up 2 compiler warnings on OS/2.
784 * Revision 1.41 2001/09/16 17:05:14 jongfoster
785 * Removing unused #include showarg.h
787 * Revision 1.40 2001/09/16 15:41:45 jongfoster
788 * Fixing signed/unsigned comparison warning.
790 * Revision 1.39 2001/09/16 13:21:27 jongfoster
791 * Changes to use new list functions.
793 * Revision 1.38 2001/09/16 13:01:46 jongfoster
794 * Removing redundant function call that zeroed zalloc()'d memory.
796 * Revision 1.37 2001/09/10 11:12:24 oes
797 * Deleted unused variable
799 * Revision 1.36 2001/09/10 10:56:15 oes
800 * Silenced compiler warnings
802 * Revision 1.35 2001/07/31 14:44:22 oes
803 * Deleted unused size parameter from filter_popups()
805 * Revision 1.34 2001/07/30 22:08:36 jongfoster
806 * Tidying up #defines:
807 * - All feature #defines are now of the form FEATURE_xxx
808 * - Permanently turned off WIN_GUI_EDIT
809 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
811 * Revision 1.33 2001/07/29 19:32:00 jongfoster
812 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
814 * Revision 1.32 2001/07/29 18:47:05 jongfoster
815 * Adding missing #include "loadcfg.h"
817 * Revision 1.31 2001/07/29 12:17:48 oes
818 * Applied pthread fix by Paul Lieverse
820 * Revision 1.30 2001/07/25 22:57:13 jongfoster
821 * __BEOS__ no longer overrides FEATURE_PTHREAD.
822 * This is because FEATURE_PTHREAD will soon be widely used, so I
823 * want to keep it simple.
825 * Revision 1.29 2001/07/24 12:47:06 oes
826 * Applied BeOS support update by Eugenia
828 * Revision 1.28 2001/07/23 13:26:12 oes
829 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
831 * Revision 1.27 2001/07/19 19:09:47 haroon
832 * - Added code to take care of the situation where while processing the first
833 * server response (which includes the server header), after finding the end
834 * of the headers we were not looking past the end of the headers for
835 * content modification. I enabled it for filter_popups.
836 * Someone else should look to see if other similar operations should be
837 * done to the discarded portion of the buffer.
839 * Note 2001/07/20: No, the other content modification mechanisms will process
840 * the whole iob later anyway. --oes
842 * Revision 1.26 2001/07/18 12:31:36 oes
845 * Revision 1.25 2001/07/15 19:43:49 jongfoster
846 * Supports POSIX threads.
847 * Also removed some unused #includes.
849 * Revision 1.24 2001/07/13 14:00:40 oes
850 * - Generic content modification scheme:
851 * Each feature has its own applicability flag that is set
852 * from csp->action->flags.
853 * Replaced the "filtering" int flag , by a function pointer
854 * "content_filter" to the function that will do the content
855 * modification. If it is != NULL, the document will be buffered
856 * and processed through *content_filter, which must set
857 * csp->content_length and return a modified copy of the body
858 * or return NULL (on failiure).
859 * - Changed csp->is_text to the more generic bitmap csp->content_type
860 * which can currently take the valued CT_TEXT or CT_GIF
862 * - Removed all #ifdef PCRS
864 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
865 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
866 * the `block_acl' call. This prevents a compilation error when the user
867 * does not wish to use the "ACL" feature.
869 * Revision 1.22 2001/06/29 21:45:41 oes
870 * Indentation, CRLF->LF, Tab-> Space
872 * Revision 1.21 2001/06/29 13:29:36 oes
873 * - Cleaned up, improved comments
874 * - Unified all possible interceptors (CGI,
875 * block, trust, fast_redirect) in one
876 * place, with one (CGI) answer generation
877 * mechansim. Much clearer now.
878 * - Removed the GIF image generation, which
879 * is now done in filters.c:block_url()
880 * - Made error conditions like domain lookup
881 * failiure or (various) problems while talking
882 * to the server use cgi.c:error_response()
883 * instead of generating HTML/HTTP in chat() (yuck!)
884 * - Removed logentry from cancelled commit
886 * Revision 1.20 2001/06/09 10:55:28 jongfoster
887 * Changing BUFSIZ ==> BUFFER_SIZE
889 * Revision 1.19 2001/06/07 23:12:52 jongfoster
890 * Replacing function pointer in struct gateway with a directly
891 * called function forwarded_connect().
892 * Replacing struct gateway with struct forward_spec
894 * Revision 1.18 2001/06/03 19:12:16 oes
895 * introduced new cgi handling
897 * Revision 1.17 2001/06/01 20:07:23 jongfoster
898 * Now uses action +image-blocker{} rather than config->tinygif
900 * Revision 1.16 2001/06/01 18:49:17 jongfoster
901 * Replaced "list_share" with "list" - the tiny memory gain was not
902 * worth the extra complexity.
904 * Revision 1.15 2001/05/31 21:24:47 jongfoster
905 * Changed "permission" to "action" throughout.
906 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
907 * Moved vanilla wafer check into chat(), since we must now
908 * decide whether or not to add it based on the URL.
910 * Revision 1.14 2001/05/29 20:14:01 joergs
911 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
914 * Revision 1.13 2001/05/29 09:50:24 jongfoster
915 * Unified blocklist/imagelist/permissionslist.
916 * File format is still under discussion, but the internal changes
919 * Also modified interceptor behaviour:
920 * - We now intercept all URLs beginning with one of the following
921 * prefixes (and *only* these prefixes):
923 * * http://ijbswa.sf.net/config/
924 * * http://ijbswa.sourceforge.net/config/
925 * - New interceptors "home page" - go to http://i.j.b/ to see it.
926 * - Internal changes so that intercepted and fast redirect pages
927 * are not replaced with an image.
928 * - Interceptors now have the option to send a binary page direct
929 * to the client. (i.e. ijb-send-banner uses this)
930 * - Implemented show-url-info interceptor. (Which is why I needed
931 * the above interceptors changes - a typical URL is
932 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
933 * The previous mechanism would not have intercepted that, and
934 * if it had been intercepted then it then it would have replaced
937 * Revision 1.12 2001/05/27 22:17:04 oes
939 * - re_process_buffer no longer writes the modified buffer
940 * to the client, which was very ugly. It now returns the
941 * buffer, which it is then written by chat.
943 * - content_length now adjusts the Content-Length: header
944 * for modified documents rather than crunch()ing it.
945 * (Length info in csp->content_length, which is 0 for
946 * unmodified documents)
948 * - For this to work, sed() is called twice when filtering.
950 * Revision 1.11 2001/05/26 17:27:53 jongfoster
951 * Added support for CLF and fixed LOG_LEVEL_LOG.
952 * Also did CRLF->LF fix of my previous patch.
954 * Revision 1.10 2001/05/26 15:26:15 jongfoster
955 * ACL feature now provides more security by immediately dropping
956 * connections from untrusted hosts.
958 * Revision 1.9 2001/05/26 00:28:36 jongfoster
959 * Automatic reloading of config file.
960 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
961 * Most of the global variables have been moved to a new
962 * struct configuration_spec, accessed through csp->config->globalname
963 * Most of the globals remaining are used by the Win32 GUI.
965 * Revision 1.8 2001/05/25 22:43:18 jongfoster
966 * Fixing minor memory leak and buffer overflow.
968 * Revision 1.7 2001/05/25 22:34:30 jongfoster
971 * Revision 1.6 2001/05/23 00:13:58 joergs
972 * AmigaOS support fixed.
974 * Revision 1.5 2001/05/22 18:46:04 oes
976 * - Enabled filtering banners by size rather than URL
977 * by adding patterns that replace all standard banner
978 * sizes with the "Junkbuster" gif to the re_filterfile
980 * - Enabled filtering WebBugs by providing a pattern
981 * which kills all 1x1 images
983 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
984 * which is selected by the (nonstandard and therefore
985 * capital) letter 'U' in the option string.
986 * It causes the quantifiers to be ungreedy by default.
987 * Appending a ? turns back to greedy (!).
989 * - Added a new interceptor ijb-send-banner, which
990 * sends back the "Junkbuster" gif. Without imagelist or
991 * MSIE detection support, or if tinygif = 1, or the
992 * URL isn't recognized as an imageurl, a lame HTML
993 * explanation is sent instead.
995 * - Added new feature, which permits blocking remote
996 * script redirects and firing back a local redirect
998 * The feature is conditionally compiled, i.e. it
999 * can be disabled with --disable-fast-redirects,
1000 * plus it must be activated by a "fast-redirects"
1001 * line in the config file, has its own log level
1002 * and of course wants to be displayed by show-proxy-args
1003 * Note: Boy, all the #ifdefs in 1001 locations and
1004 * all the fumbling with configure.in and acconfig.h
1005 * were *way* more work than the feature itself :-(
1007 * - Because a generic redirect template was needed for
1008 * this, tinygif = 3 now uses the same.
1010 * - Moved GIFs, and other static HTTP response templates
1013 * - Some minor fixes
1015 * - Removed some >400 CRs again (Jon, you really worked
1018 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1019 * Made failure to bind() a fatal error.
1021 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1022 * Version 2.9.4 checkin.
1023 * - Merged popupfile and cookiefile, and added control over PCRS
1024 * filtering, in new "permissionsfile".
1025 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1026 * file error you now get a message box (in the Win32 GUI) rather
1027 * than the program exiting with no explanation.
1028 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1030 * - Removed tabs from "config"
1031 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1032 * - Bumped up version number.
1034 * Revision 1.2 2001/05/17 22:34:44 oes
1035 * - Added hint on GIF char array generation to jcc.c
1036 * - Cleaned CRLF's from the sources and related files
1037 * - Repaired logging for REF and FRC
1039 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1040 * Initial import of version 2.9.3 source tree
1043 *********************************************************************/
1049 #include <sys/types.h>
1058 # ifndef FEATURE_PTHREAD
1062 # include <windows.h>
1063 # include <process.h>
1064 # endif /* ndef FEATURE_PTHREAD */
1067 # ifndef _WIN_CONSOLE
1068 # include "w32log.h"
1069 # endif /* ndef _WIN_CONSOLE */
1070 # include "w32svrapi.h"
1072 #else /* ifndef _WIN32 */
1074 # if !defined (__OS2__)
1075 # include <unistd.h>
1076 # include <sys/wait.h>
1077 # endif /* ndef __OS2__ */
1078 # include <sys/time.h>
1079 # include <sys/stat.h>
1080 # include <sys/ioctl.h>
1083 #include <sys/termios.h>
1091 # include <signal.h>
1094 # include <socket.h> /* BeOS has select() for sockets only. */
1095 # include <OS.h> /* declarations for threads and stuff. */
1098 # if defined(__EMX__) || defined(__OS2__)
1099 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1104 #define bzero(B,N) memset(B,0x00,n)
1108 # include <select.h>
1113 #include "project.h"
1116 #include "filters.h"
1117 #include "loaders.h"
1118 #include "parsers.h"
1119 #include "miscutil.h"
1121 #include "jbsockets.h"
1122 #include "gateway.h"
1123 #include "actions.h"
1125 #include "loadcfg.h"
1126 #include "urlmatch.h"
1128 const char jcc_h_rcs[] = JCC_H_VERSION;
1129 const char project_h_rcs[] = PROJECT_H_VERSION;
1132 struct client_state clients[1];
1133 struct file_list files[1];
1135 #ifdef FEATURE_STATISTICS
1136 int urls_read = 0; /* total nr of urls read inc rejected */
1137 int urls_rejected = 0; /* total nr of urls rejected */
1138 #endif /* def FEATURE_STATISTICS */
1140 #ifdef FEATURE_GRACEFUL_TERMINATION
1141 int g_terminate = 0;
1144 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1145 static void sig_handler(int the_signal);
1147 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1148 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1149 static jb_err get_server_headers(struct client_state *csp);
1150 static const char *crunch_reason(const struct http_response *rsp);
1151 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1153 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1155 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1156 static jb_err change_request_destination(struct client_state *csp);
1157 static void chat(struct client_state *csp);
1158 static void serve(struct client_state *csp);
1159 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1160 static void usage(const char *myname);
1162 static void initialize_mutexes(void);
1163 static jb_socket bind_port_helper(struct configuration_spec *config);
1164 static void listen_loop(void);
1167 void serve(struct client_state *csp);
1168 #else /* ifndef AMIGA */
1169 static void serve(struct client_state *csp);
1170 #endif /* def AMIGA */
1173 static int32 server_thread(void *data);
1174 #endif /* def __BEOS__ */
1177 #define sleep(N) Sleep(((N) * 1000))
1181 #define sleep(N) DosSleep(((N) * 100))
1184 #ifdef MUTEX_LOCKS_AVAILABLE
1186 * XXX: Does the locking stuff really belong in this file?
1188 privoxy_mutex_t log_mutex;
1189 privoxy_mutex_t log_init_mutex;
1190 privoxy_mutex_t connection_reuse_mutex;
1192 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1193 privoxy_mutex_t resolver_mutex;
1194 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1196 #ifndef HAVE_GMTIME_R
1197 privoxy_mutex_t gmtime_mutex;
1198 #endif /* ndef HAVE_GMTIME_R */
1200 #ifndef HAVE_LOCALTIME_R
1201 privoxy_mutex_t localtime_mutex;
1202 #endif /* ndef HAVE_GMTIME_R */
1205 privoxy_mutex_t rand_mutex;
1206 #endif /* ndef HAVE_RANDOM */
1208 #endif /* def MUTEX_LOCKS_AVAILABLE */
1211 const char *basedir = NULL;
1212 const char *pidfile = NULL;
1213 static int received_hup_signal = 0;
1214 #endif /* defined unix */
1216 /* HTTP snipplets. */
1217 static const char CSUCCEED[] =
1218 "HTTP/1.0 200 Connection established\r\n"
1219 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1221 static const char CHEADER[] =
1222 "HTTP/1.0 400 Invalid header received from client\r\n"
1223 "Proxy-Agent: Privoxy " VERSION "\r\n"
1224 "Content-Type: text/plain\r\n"
1225 "Connection: close\r\n\r\n"
1226 "Invalid header received from client.\r\n";
1228 static const char FTP_RESPONSE[] =
1229 "HTTP/1.0 400 Invalid request received from client\r\n"
1230 "Content-Type: text/plain\r\n"
1231 "Connection: close\r\n\r\n"
1232 "Invalid request. Privoxy doesn't support FTP.\r\n";
1234 static const char GOPHER_RESPONSE[] =
1235 "HTTP/1.0 400 Invalid request received from client\r\n"
1236 "Content-Type: text/plain\r\n"
1237 "Connection: close\r\n\r\n"
1238 "Invalid request. Privoxy doesn't support gopher.\r\n";
1240 /* XXX: should be a template */
1241 static const char MISSING_DESTINATION_RESPONSE[] =
1242 "HTTP/1.0 400 Bad request received from client\r\n"
1243 "Proxy-Agent: Privoxy " VERSION "\r\n"
1244 "Content-Type: text/plain\r\n"
1245 "Connection: close\r\n\r\n"
1246 "Bad request. Privoxy was unable to extract the destination.\r\n";
1248 /* XXX: should be a template */
1249 static const char NO_SERVER_DATA_RESPONSE[] =
1250 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1251 "Proxy-Agent: Privoxy " VERSION "\r\n"
1252 "Content-Type: text/plain\r\n"
1253 "Connection: close\r\n\r\n"
1254 "Empty server or forwarder response.\r\n"
1255 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1257 /* XXX: should be a template */
1258 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1259 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1260 "Proxy-Agent: Privoxy " VERSION "\r\n"
1261 "Content-Type: text/plain\r\n"
1262 "Connection: close\r\n\r\n"
1263 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1266 /* XXX: should be a template */
1267 static const char NULL_BYTE_RESPONSE[] =
1268 "HTTP/1.0 400 Bad request received from client\r\n"
1269 "Proxy-Agent: Privoxy " VERSION "\r\n"
1270 "Content-Type: text/plain\r\n"
1271 "Connection: close\r\n\r\n"
1272 "Bad request. Null byte(s) before end of request.\r\n";
1275 /* XXX: should be a template */
1276 static const char MESSED_UP_REQUEST_RESPONSE[] =
1277 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1278 "Proxy-Agent: Privoxy " VERSION "\r\n"
1279 "Content-Type: text/plain\r\n"
1280 "Connection: close\r\n\r\n"
1281 "Bad request. Messed up with header filters.\r\n";
1283 /* A function to crunch a response */
1284 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1286 /* Crunch function flags */
1287 #define CF_NO_FLAGS 0
1288 /* Cruncher applies to forced requests as well */
1289 #define CF_IGNORE_FORCE 1
1290 /* Crunched requests are counted for the block statistics */
1291 #define CF_COUNT_AS_REJECT 2
1293 /* A crunch function and its flags */
1296 const crunch_func_ptr cruncher;
1300 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1302 /* Complete list of cruncher functions */
1303 static const struct cruncher crunchers_all[] = {
1304 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1305 { block_url, CF_COUNT_AS_REJECT },
1306 #ifdef FEATURE_TRUST
1307 { trust_url, CF_COUNT_AS_REJECT },
1308 #endif /* def FEATURE_TRUST */
1309 { redirect_url, CF_NO_FLAGS },
1310 { dispatch_cgi, CF_IGNORE_FORCE},
1314 /* Light version, used after tags are applied */
1315 static const struct cruncher crunchers_light[] = {
1316 { block_url, CF_COUNT_AS_REJECT },
1317 { redirect_url, CF_NO_FLAGS },
1323 * XXX: Don't we really mean
1329 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1330 /*********************************************************************
1332 * Function : sig_handler
1334 * Description : Signal handler for different signals.
1335 * Exit gracefully on TERM and INT
1336 * or set a flag that will cause the errlog
1337 * to be reopened by the main thread on HUP.
1340 * 1 : the_signal = the signal cause this function to call
1344 *********************************************************************/
1345 static void sig_handler(int the_signal)
1351 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1363 received_hup_signal = 1;
1369 * We shouldn't be here, unless we catch signals
1370 * in main() that we can't handle here!
1372 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1380 /*********************************************************************
1382 * Function : client_protocol_is_unsupported
1384 * Description : Checks if the client used a known unsupported
1385 * protocol and deals with it by sending an error
1389 * 1 : csp = Current client state (buffers, headers, etc...)
1390 * 2 : req = the first request line send by the client
1392 * Returns : TRUE if an error response has been generated, or
1393 * FALSE if the request doesn't look invalid.
1395 *********************************************************************/
1396 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1399 * If it's a FTP or gopher request, we don't support it.
1401 * These checks are better than nothing, but they might
1402 * not work in all configurations and some clients might
1403 * have problems digesting the answer.
1405 * They should, however, never cause more problems than
1406 * Privoxy's old behaviour (returning the misleading HTML
1409 * "Could not resolve http://(ftp|gopher)://example.org").
1411 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1413 const char *response = NULL;
1414 const char *protocol = NULL;
1416 if (!strncmpic(req, "GET ftp://", 10))
1418 response = FTP_RESPONSE;
1423 response = GOPHER_RESPONSE;
1424 protocol = "GOPHER";
1426 log_error(LOG_LEVEL_ERROR,
1427 "%s tried to use Privoxy as %s proxy: %s",
1428 csp->ip_addr_str, protocol, req);
1429 log_error(LOG_LEVEL_CLF,
1430 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1432 write_socket(csp->cfd, response, strlen(response));
1441 /*********************************************************************
1443 * Function : get_request_destination_elsewhere
1445 * Description : If the client's request was redirected into
1446 * Privoxy without the client's knowledge,
1447 * the request line lacks the destination host.
1449 * This function tries to get it elsewhere,
1450 * provided accept-intercepted-requests is enabled.
1452 * "Elsewhere" currently only means "Host: header",
1453 * but in the future we may ask the redirecting
1454 * packet filter to look the destination up.
1456 * If the destination stays unknown, an error
1457 * response is send to the client and headers
1458 * are freed so that chat() can return directly.
1461 * 1 : csp = Current client state (buffers, headers, etc...)
1462 * 2 : headers = a header list
1464 * Returns : JB_ERR_OK if the destination is now known, or
1465 * JB_ERR_PARSE if it isn't.
1467 *********************************************************************/
1468 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1472 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1474 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1475 " Privoxy isn't configured to accept intercepted requests.",
1476 csp->ip_addr_str, csp->http->cmd);
1477 /* XXX: Use correct size */
1478 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1479 csp->ip_addr_str, csp->http->cmd);
1481 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1482 destroy_list(headers);
1484 return JB_ERR_PARSE;
1486 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1488 /* Split the domain we just got for pattern matching */
1489 init_domain_components(csp->http);
1495 /* We can't work without destination. Go spread the news.*/
1497 req = list_to_text(headers);
1499 /* XXX: Use correct size */
1500 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1501 csp->ip_addr_str, csp->http->cmd);
1502 log_error(LOG_LEVEL_ERROR,
1503 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1504 csp->ip_addr_str, csp->http->cmd, req);
1507 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1508 destroy_list(headers);
1510 return JB_ERR_PARSE;
1513 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1514 * to get the destination IP address, use it as host directly
1515 * or do a reverse DNS lookup first.
1520 /*********************************************************************
1522 * Function : get_server_headers
1524 * Description : Parses server headers in iob and fills them
1525 * into csp->headers so that they can later be
1529 * 1 : csp = Current client state (buffers, headers, etc...)
1531 * Returns : JB_ERR_OK if everything went fine, or
1532 * JB_ERR_PARSE if the headers were incomplete.
1534 *********************************************************************/
1535 static jb_err get_server_headers(struct client_state *csp)
1537 int continue_hack_in_da_house = 0;
1540 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1545 * continue hack in da house. Ignore the ending of
1546 * this head and continue enlisting header lines.
1547 * The reason is described below.
1549 enlist(csp->headers, "");
1550 continue_hack_in_da_house = 0;
1553 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1556 * It's a bodyless continue response, don't
1557 * stop header parsing after reaching its end.
1559 * As a result Privoxy will concatenate the
1560 * next response's head and parse and deliver
1561 * the headers as if they belonged to one request.
1563 * The client will separate them because of the
1564 * empty line between them.
1566 * XXX: What we're doing here is clearly against
1567 * the intended purpose of the continue header,
1568 * and under some conditions (HTTP/1.0 client request)
1569 * it's a standard violation.
1571 * Anyway, "sort of against the spec" is preferable
1572 * to "always getting confused by Continue responses"
1573 * (Privoxy's behaviour before this hack was added)
1575 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1576 continue_hack_in_da_house = 1;
1578 else if (*header == '\0')
1581 * If the header is empty, but the Continue hack
1582 * isn't active, we can assume that we reached the
1583 * end of the buffer before we hit the end of the
1586 * Inform the caller an let it decide how to handle it.
1588 return JB_ERR_PARSE;
1591 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1594 * XXX: Should we quit the request and return a
1595 * out of memory error page instead?
1597 log_error(LOG_LEVEL_ERROR,
1598 "Out of memory while enlisting server headers. %s lost.",
1608 /*********************************************************************
1610 * Function : crunch_reason
1612 * Description : Translates the crunch reason code into a string.
1615 * 1 : rsp = a http_response
1617 * Returns : A string with the crunch reason or an error description.
1619 *********************************************************************/
1620 static const char *crunch_reason(const struct http_response *rsp)
1622 char * reason = NULL;
1624 assert(rsp != NULL);
1627 return "Internal error while searching for crunch reason";
1630 switch (rsp->reason)
1632 case RSP_REASON_UNSUPPORTED:
1633 reason = "Unsupported HTTP feature";
1635 case RSP_REASON_BLOCKED:
1638 case RSP_REASON_UNTRUSTED:
1639 reason = "Untrusted";
1641 case RSP_REASON_REDIRECTED:
1642 reason = "Redirected";
1644 case RSP_REASON_CGI_CALL:
1645 reason = "CGI Call";
1647 case RSP_REASON_NO_SUCH_DOMAIN:
1648 reason = "DNS failure";
1650 case RSP_REASON_FORWARDING_FAILED:
1651 reason = "Forwarding failed";
1653 case RSP_REASON_CONNECT_FAILED:
1654 reason = "Connection failure";
1656 case RSP_REASON_OUT_OF_MEMORY:
1657 reason = "Out of memory (may mask other reasons)";
1660 reason = "No reason recorded";
1668 /*********************************************************************
1670 * Function : send_crunch_response
1672 * Description : Delivers already prepared response for
1673 * intercepted requests, logs the interception
1674 * and frees the response.
1677 * 1 : csp = Current client state (buffers, headers, etc...)
1678 * 1 : rsp = Fully prepared response. Will be freed on exit.
1680 * Returns : Nothing.
1682 *********************************************************************/
1683 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1685 const struct http_request *http = csp->http;
1686 char status_code[4];
1688 assert(rsp != NULL);
1689 assert(rsp->head != NULL);
1694 * Not supposed to happen. If it does
1695 * anyway, treat it as an unknown error.
1697 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1698 /* return code doesn't matter */
1703 /* If rsp is still NULL, we have serious internal problems. */
1704 log_error(LOG_LEVEL_FATAL,
1705 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1709 * Extract the status code from the actual head
1710 * that was send to the client. It is the only
1711 * way to get it right for all requests, including
1712 * the fixed ones for out-of-memory problems.
1714 * A head starts like this: 'HTTP/1.1 200...'
1718 status_code[0] = rsp->head[9];
1719 status_code[1] = rsp->head[10];
1720 status_code[2] = rsp->head[11];
1721 status_code[3] = '\0';
1723 /* Write the answer to the client */
1724 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1725 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1727 /* There is nothing we can do about it. */
1728 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1731 /* Log that the request was crunched and why. */
1732 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1733 http->hostport, http->path, crunch_reason(rsp));
1734 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1735 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1737 /* Clean up and return */
1738 if (cgi_error_memory() != rsp)
1740 free_http_response(rsp);
1747 /*********************************************************************
1749 * Function : request_contains_null_bytes
1751 * Description : Checks for NULL bytes in the request and sends
1752 * an error message to the client if any were found.
1754 * XXX: currently not used, see comment in chat().
1757 * 1 : csp = Current client state (buffers, headers, etc...)
1758 * 2 : buf = Data from the client's request to check.
1759 * 3 : len = The data length.
1761 * Returns : TRUE if the request contained one or more NULL bytes, or
1764 *********************************************************************/
1765 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1767 size_t c_len; /* Request lenght when treated as C string */
1769 c_len = strlen(buf);
1774 * Null byte(s) found. Log the request,
1775 * return an error response and hang up.
1777 size_t tmp_len = c_len;
1782 * Replace NULL byte(s) with '°' characters
1783 * so the request can be logged as string.
1784 * XXX: Is there a better replacement character?
1787 tmp_len += strlen(buf+tmp_len);
1788 } while (tmp_len < len);
1790 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1791 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1792 log_error(LOG_LEVEL_HEADER,
1793 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1795 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1797 /* XXX: Log correct size */
1798 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1808 /*********************************************************************
1810 * Function : crunch_response_triggered
1812 * Description : Checks if the request has to be crunched,
1813 * and delivers the crunch response if necessary.
1816 * 1 : csp = Current client state (buffers, headers, etc...)
1817 * 2 : crunchers = list of cruncher functions to run
1819 * Returns : TRUE if the request was answered with a crunch response
1822 *********************************************************************/
1823 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1825 struct http_response *rsp = NULL;
1826 const struct cruncher *c;
1829 * If CGI request crunching is disabled,
1830 * check the CGI dispatcher out of order to
1831 * prevent unintentional blocks or redirects.
1833 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1834 && (NULL != (rsp = dispatch_cgi(csp))))
1836 /* Deliver, log and free the interception response. */
1837 send_crunch_response(csp, rsp);
1841 for (c = crunchers; c->cruncher != NULL; c++)
1844 * Check the cruncher if either Privoxy is toggled
1845 * on and the request isn't forced, or if the cruncher
1846 * applies to forced requests as well.
1848 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1849 !(csp->flags & CSP_FLAG_FORCED)) ||
1850 (c->flags & CF_IGNORE_FORCE))
1852 rsp = c->cruncher(csp);
1855 /* Deliver, log and free the interception response. */
1856 send_crunch_response(csp, rsp);
1857 #ifdef FEATURE_STATISTICS
1858 if (c->flags & CF_COUNT_AS_REJECT)
1860 csp->flags |= CSP_FLAG_REJECTED;
1862 #endif /* def FEATURE_STATISTICS */
1873 /*********************************************************************
1875 * Function : build_request_line
1877 * Description : Builds the HTTP request line.
1879 * If a HTTP forwarder is used it expects the whole URL,
1880 * web servers only get the path.
1883 * 1 : csp = Current client state (buffers, headers, etc...)
1884 * 2 : fwd = The forwarding spec used for the request
1885 * XXX: Should use http->fwd instead.
1886 * 3 : request_line = The old request line which will be replaced.
1888 * Returns : Nothing. Terminates in case of memory problems.
1890 *********************************************************************/
1891 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1893 struct http_request *http = csp->http;
1895 assert(http->ssl == 0);
1898 * Downgrade http version from 1.1 to 1.0
1899 * if +downgrade action applies.
1901 if ( (csp->action->flags & ACTION_DOWNGRADE)
1902 && (!strcmpic(http->ver, "HTTP/1.1")))
1905 http->ver = strdup("HTTP/1.0");
1907 if (http->ver == NULL)
1909 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1914 * Rebuild the request line.
1916 freez(*request_line);
1917 *request_line = strdup(http->gpc);
1918 string_append(request_line, " ");
1920 if (fwd->forward_host)
1922 string_append(request_line, http->url);
1926 string_append(request_line, http->path);
1928 string_append(request_line, " ");
1929 string_append(request_line, http->ver);
1931 if (*request_line == NULL)
1933 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1935 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1939 /*********************************************************************
1941 * Function : change_request_destination
1943 * Description : Parse a (rewritten) request line and regenerate
1944 * the http request data.
1947 * 1 : csp = Current client state (buffers, headers, etc...)
1949 * Returns : Forwards the parse_http_request() return code.
1950 * Terminates in case of memory problems.
1952 *********************************************************************/
1953 static jb_err change_request_destination(struct client_state *csp)
1955 struct http_request *http = csp->http;
1958 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1959 free_http_request(http);
1960 err = parse_http_request(csp->headers->first->str, http, csp);
1961 if (JB_ERR_OK != err)
1963 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1964 jb_err_to_string(err));
1968 /* XXX: ocmd is a misleading name */
1969 http->ocmd = strdup(http->cmd);
1970 if (http->ocmd == NULL)
1972 log_error(LOG_LEVEL_FATAL,
1973 "Out of memory copying rewritten HTTP request line");
1981 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
1982 /*********************************************************************
1984 * Function : server_response_is_complete
1986 * Description : Determines whether we should stop reading
1987 * from the server socket.
1990 * 1 : csp = Current client state (buffers, headers, etc...)
1991 * 2 : content_length = Length of content received so far.
1993 * Returns : TRUE if the response is complete,
1996 *********************************************************************/
1997 static int server_response_is_complete(struct client_state *csp, size_t content_length)
1999 int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2001 if (!strcmpic(csp->http->gpc, "HEAD"))
2004 * "HEAD" implies no body, we are thus expecting
2005 * no content. XXX: incomplete "list" of methods?
2007 log_error(LOG_LEVEL_INFO, "Method %s implies no body.", csp->http->gpc);
2008 csp->expected_content_length = 0;
2009 content_length_known = TRUE;
2012 if (csp->http->status == 304)
2015 * Expect no body. XXX: incomplete "list" of status codes?
2017 log_error(LOG_LEVEL_INFO, "Status code %d implies no body.", csp->http->status);
2018 csp->expected_content_length = 0;
2019 content_length_known = TRUE;
2022 return (content_length_known && ((0 == csp->expected_content_length)
2023 || (csp->expected_content_length <= content_length)));
2025 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2028 /*********************************************************************
2032 * Description : Once a connection to the client has been accepted,
2033 * this function is called (via serve()) to handle the
2034 * main business of the communication. When this
2035 * function returns, the caller must close the client
2038 * FIXME: chat is nearly thousand lines long.
2042 * 1 : csp = Current client state (buffers, headers, etc...)
2044 * Returns : Nothing.
2046 *********************************************************************/
2047 static void chat(struct client_state *csp)
2049 char buf[BUFFER_SIZE];
2057 int ms_iis5_hack = 0;
2058 size_t byte_count = 0;
2059 int forwarded_connect_retries = 0;
2060 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2061 const struct forward_spec * fwd;
2062 struct http_request *http;
2063 int len; /* for buffer sizes (and negative error codes) */
2066 /* Function that does the content filtering for the current request */
2067 filter_function_ptr content_filter = NULL;
2069 /* Skeleton for HTTP response, if we should intercept the request */
2070 struct http_response *rsp;
2072 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2073 struct list header_list;
2074 struct list *headers = &header_list;
2078 memset(buf, 0, sizeof(buf));
2081 * Read the client's request. Note that since we're not using select() we
2082 * could get blocked here if a client connected, then didn't say anything!
2087 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2089 if (len <= 0) break; /* error! */
2092 * If there is no memory left for buffering the
2093 * request, there is nothing we can do but hang up
2095 if (add_to_iob(csp, buf, len))
2100 req = get_header(csp->iob);
2102 } while ((NULL != req) && ('\0' == *req));
2104 if ((NULL != req) && ('\0' != *req))
2106 /* Request received. Validate and parse it. */
2110 * XXX: Temporary disabled to prevent problems
2111 * with POST requests whose bodies are allowed to
2112 * contain NULL bytes. BR#1730105.
2114 * The main purpose of this check is to properly
2115 * log stuff like BitTorrent traffic and other junk
2116 * that hits public proxies. It's not required for
2117 * Privoxy to functions as those requests are discarded
2120 * It probably should be rewritten to only check
2121 * the head of the request. Another option would
2122 * be to let all POST requests pass, although that
2123 * may not be good enough.
2125 if (request_contains_null_bytes(csp, buf, len))
2127 /* NULL bytes found and dealt with, just hang up. */
2132 /* Does the request line look invalid? */
2133 if (client_protocol_is_unsupported(csp, req))
2136 * Yes. The request has already been
2137 * answered with a error response, the buffers
2138 * were freed and we're done with chatting.
2143 #ifdef FEATURE_FORCE_LOAD
2145 * If this request contains the FORCE_PREFIX and blocks
2146 * aren't enforced, get rid of it and set the force flag.
2148 if (strstr(req, FORCE_PREFIX))
2150 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2152 log_error(LOG_LEVEL_FORCE,
2153 "Ignored force prefix in request: \"%s\".", req);
2157 strclean(req, FORCE_PREFIX);
2158 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2159 csp->flags |= CSP_FLAG_FORCED;
2163 #endif /* def FEATURE_FORCE_LOAD */
2164 err = parse_http_request(req, http, csp);
2165 if (JB_ERR_OK != err)
2167 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2173 if (http->cmd == NULL)
2175 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2176 /* XXX: Use correct size */
2177 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2178 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2180 free_http_request(http);
2184 /* grab the rest of the client's headers */
2188 p = get_header(csp->iob);
2192 /* There are no additional headers to read. */
2199 * We didn't receive a complete header
2200 * line yet, get the rest of it.
2202 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2205 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2206 destroy_list(headers);
2210 if (add_to_iob(csp, buf, len))
2213 * If there is no memory left for buffering the
2214 * request, there is nothing we can do but hang up
2216 destroy_list(headers);
2223 * We were able to read a complete
2224 * header and can finaly enlist it.
2231 if (http->host == NULL)
2234 * If we still don't know the request destination,
2235 * the request is invalid or the client uses
2236 * Privoxy without its knowledge.
2238 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2241 * Our attempts to get the request destination
2242 * elsewhere failed or Privoxy is configured
2243 * to only accept proxy requests.
2245 * An error response has already been send
2246 * and we're done here.
2253 * Determine the actions for this URL
2255 #ifdef FEATURE_TOGGLE
2256 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2258 /* Most compatible set of actions (i.e. none) */
2259 init_current_action(csp->action);
2262 #endif /* ndef FEATURE_TOGGLE */
2264 get_url_actions(csp, http);
2268 * Save a copy of the original request for logging
2270 http->ocmd = strdup(http->cmd);
2272 if (http->ocmd == NULL)
2274 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2277 enlist(csp->headers, http->cmd);
2279 /* Append the previously read headers */
2280 list_append_list_unique(csp->headers, headers);
2281 destroy_list(headers);
2283 err = sed(csp, FILTER_CLIENT_HEADERS);
2284 if (JB_ERR_OK != err)
2286 assert(err == JB_ERR_PARSE);
2287 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2289 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2291 /* Check request line for rewrites. */
2292 if ((NULL == csp->headers->first->str)
2293 || (strcmp(http->cmd, csp->headers->first->str) &&
2294 (JB_ERR_OK != change_request_destination(csp))))
2297 * A header filter broke the request line - bail out.
2299 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2300 /* XXX: Use correct size */
2301 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2302 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2304 free_http_request(http);
2308 /* decide how to route the HTTP request */
2309 fwd = forward_url(csp, http);
2312 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2313 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2317 * build the http request to send to the server
2318 * we have to do one of the following:
2320 * create = use the original HTTP request to create a new
2321 * HTTP request that has either the path component
2322 * without the http://domainspec (w/path) or the
2323 * full orininal URL (w/url)
2324 * Note that the path and/or the HTTP version may
2325 * have been altered by now.
2327 * connect = Open a socket to the host:port of the server
2328 * and short-circuit server and client socket.
2330 * pass = Pass the request unchanged if forwarding a CONNECT
2331 * request to a parent proxy. Note that we'll be sending
2332 * the CFAIL message ourselves if connecting to the parent
2333 * fails, but we won't send a CSUCCEED message if it works,
2334 * since that would result in a double message (ours and the
2335 * parent's). After sending the request to the parent, we simply
2338 * here's the matrix:
2341 * +--------+--------+
2343 * 0 | create | connect|
2345 * Forwarding +--------+--------+
2347 * 1 | create | pass |
2349 * +--------+--------+
2353 if (http->ssl && connect_port_is_forbidden(csp))
2355 const char *acceptable_connect_ports =
2356 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2357 assert(NULL != acceptable_connect_ports);
2358 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2359 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2360 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2361 csp->action->flags |= ACTION_BLOCK;
2367 freez(csp->headers->first->str);
2368 build_request_line(csp, fwd, &csp->headers->first->str);
2371 hdr = list_to_text(csp->headers);
2374 /* FIXME Should handle error properly */
2375 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2379 * We have a request. Check if one of the crunchers wants it.
2381 if (crunch_response_triggered(csp, crunchers_all))
2384 * Yes. The client got the crunch response
2385 * and we are done here after cleaning up.
2388 list_remove_all(csp->headers);
2394 * The headers can't be removed earlier because
2395 * they were still needed for the referrer check
2396 * in case of CGI crunches.
2398 * XXX: Would it be worth to move the referrer check
2399 * into client_referrer() and set a flag if it's trusted?
2401 list_remove_all(csp->headers);
2403 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2405 if (fwd->forward_host)
2407 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2408 fwd->forward_host, fwd->forward_port, http->hostport);
2412 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2415 /* here we connect to the server, gateway, or the forwarder */
2417 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2418 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2420 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2421 forwarded_connect_retries, http->hostport);
2424 if (csp->sfd == JB_INVALID_SOCKET)
2426 if (fwd->type != SOCKS_NONE)
2429 rsp = error_response(csp, "forwarding-failed", errno);
2431 else if (errno == EINVAL)
2433 rsp = error_response(csp, "no-such-domain", errno);
2437 rsp = error_response(csp, "connect-failed", errno);
2438 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2443 /* Write the answer to the client */
2446 send_crunch_response(csp, rsp);
2453 if (fwd->forward_host || (http->ssl == 0))
2455 /* write the client's (modified) header to the server
2456 * (along with anything else that may be in the buffer)
2459 if (write_socket(csp->sfd, hdr, strlen(hdr))
2460 || (flush_socket(csp->sfd, csp->iob) < 0))
2462 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2465 rsp = error_response(csp, "connect-failed", errno);
2469 send_crunch_response(csp, rsp);
2479 * We're running an SSL tunnel and we're not forwarding,
2480 * so just send the "connect succeeded" message to the
2481 * client, flush the rest, and get out of the way.
2483 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2491 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2493 /* we're finished with the client's header */
2496 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2498 /* pass data between the client and server
2499 * until one or the other shuts down the connection.
2508 * FD_ZERO here seems to point to an errant macro which crashes.
2509 * So do this by hand for now...
2511 memset(&rfds,0x00,sizeof(fd_set));
2515 FD_SET(csp->cfd, &rfds);
2516 FD_SET(csp->sfd, &rfds);
2518 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2519 if (server_body && server_response_is_complete(csp, byte_count))
2521 log_error(LOG_LEVEL_CONNECT,
2522 "Done reading from server. Expected content length: %d. "
2523 "Actual content length: %d. Most recently received: %d.",
2524 csp->expected_content_length, byte_count, len);
2527 * XXX: should not jump around,
2528 * chat() is complicated enough already.
2532 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2534 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2538 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2542 /* this is the body of the browser's request
2543 * just read it and write it.
2546 if (FD_ISSET(csp->cfd, &rfds))
2548 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2552 break; /* "game over, man" */
2555 if (write_socket(csp->sfd, buf, (size_t)len))
2557 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2564 * The server wants to talk. It could be the header or the body.
2565 * If `hdr' is null, then it's the header otherwise it's the body.
2566 * FIXME: Does `hdr' really mean `host'? No.
2570 if (FD_ISSET(csp->sfd, &rfds))
2573 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2577 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2579 if (http->ssl && (fwd->forward_host == NULL))
2582 * Just hang up. We already confirmed the client's CONNECT
2583 * request with status code 200 and unencrypted content is
2584 * no longer welcome.
2586 log_error(LOG_LEVEL_ERROR,
2587 "CONNECT already confirmed. Unable to tell the client about the problem.");
2590 else if (byte_count)
2593 * Just hang up. We already transmitted the original headers
2594 * and parts of the original content and therefore missed the
2595 * chance to send an error message (without risking data corruption).
2597 * XXX: we could retry with a fancy range request here.
2599 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2600 "Unable to tell the client about the problem.");
2604 rsp = error_response(csp, "connect-failed", errno);
2608 send_crunch_response(csp, rsp);
2614 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2616 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2618 /* Add a trailing zero. This lets filter_popups
2619 * use string operations.
2623 /* Normally, this would indicate that we've read
2624 * as much as the server has sent us and we can
2625 * close the client connection. However, Microsoft
2626 * in its wisdom has released IIS/5 with a bug that
2627 * prevents it from sending the trailing \r\n in
2628 * a 302 redirect header (and possibly other headers).
2629 * To work around this if we've haven't parsed
2630 * a full header we'll append a trailing \r\n
2631 * and see if this now generates a valid one.
2633 * This hack shouldn't have any impacts. If we've
2634 * already transmitted the header or if this is a
2635 * SSL connection, then we won't bother with this
2636 * hack. So we only work on partially received
2637 * headers. If we append a \r\n and this still
2638 * doesn't generate a valid header, then we won't
2639 * transmit anything to the client.
2644 if (server_body || http->ssl)
2647 * If we have been buffering up the document,
2648 * now is the time to apply content modification
2649 * and send the result to the client.
2653 p = execute_content_filter(csp, content_filter);
2655 * If the content filter fails, use the original
2656 * buffer and length.
2657 * (see p != NULL ? p : csp->iob->cur below)
2661 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2664 if (JB_ERR_OK != update_server_headers(csp))
2666 log_error(LOG_LEVEL_FATAL,
2667 "Failed to update server headers. after filtering.");
2670 hdr = list_to_text(csp->headers);
2673 /* FIXME Should handle error properly */
2674 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2677 if (write_socket(csp->cfd, hdr, strlen(hdr))
2678 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2680 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2690 break; /* "game over, man" */
2694 * This is NOT the body, so
2695 * Let's pretend the server just sent us a blank line.
2697 snprintf(buf, sizeof(buf), "\r\n");
2698 len = (int)strlen(buf);
2701 * Now, let the normal header parsing algorithm below do its
2702 * job. If it fails, we'll exit instead of continuing.
2709 * If this is an SSL connection or we're in the body
2710 * of the server document, just write it to the client,
2711 * unless we need to buffer the body for later content-filtering
2714 if (server_body || http->ssl)
2719 * If there is no memory left for buffering the content, or the buffer limit
2720 * has been reached, switch to non-filtering mode, i.e. make & write the
2721 * header, flush the iob and buf, and get out of the way.
2723 if (add_to_iob(csp, buf, len))
2728 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2730 hdr = list_to_text(csp->headers);
2734 * Memory is too tight to even generate the header.
2735 * Send our static "Out-of-memory" page.
2737 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2738 rsp = cgi_error_memory();
2739 send_crunch_response(csp, rsp);
2743 hdrlen = strlen(hdr);
2745 if (write_socket(csp->cfd, hdr, hdrlen)
2746 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2747 || (write_socket(csp->cfd, buf, (size_t)len)))
2749 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2756 * Reset the byte_count to the amount of bytes
2757 * we just flushed. len will be added a few lines below,
2758 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2760 byte_count = (size_t)flushed;
2762 content_filter = NULL;
2768 if (write_socket(csp->cfd, buf, (size_t)len))
2770 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2774 byte_count += (size_t)len;
2779 /* we're still looking for the end of the
2780 * server's header ... (does that make header
2781 * parsing an "out of body experience" ?
2785 * buffer up the data we just read. If that fails,
2786 * there's little we can do but send our static
2787 * out-of-memory page.
2789 if (add_to_iob(csp, buf, len))
2791 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2792 rsp = cgi_error_memory();
2793 send_crunch_response(csp, rsp);
2798 /* Convert iob into something sed() can digest */
2799 if (JB_ERR_PARSE == get_server_headers(csp))
2803 /* Well, we tried our MS IIS/5
2804 * hack and it didn't work.
2805 * The header is incomplete
2806 * and there isn't anything
2807 * we can do about it.
2809 log_error(LOG_LEVEL_INFO,
2810 "MS IIS5 hack didn't produce valid headers.");
2815 /* Since we have to wait for
2816 * more from the server before
2817 * we can parse the headers
2818 * we just continue here.
2824 /* Did we actually get anything? */
2825 if (NULL == csp->headers->first)
2827 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2828 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2829 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2830 free_http_request(http);
2834 assert(csp->headers->first->str);
2836 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2837 strncmpic(csp->headers->first->str, "ICY", 3))
2840 * It doesn't look like a HTTP (or Shoutcast) response:
2841 * tell the client and log the problem.
2843 if (strlen(csp->headers->first->str) > 30)
2845 csp->headers->first->str[30] = '\0';
2847 log_error(LOG_LEVEL_ERROR,
2848 "Invalid server or forwarder response. Starts with: %s",
2849 csp->headers->first->str);
2850 log_error(LOG_LEVEL_CLF,
2851 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2852 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2853 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2854 free_http_request(http);
2859 * We have now received the entire server header,
2860 * filter it and send the result to the client
2862 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2864 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2866 hdr = list_to_text(csp->headers);
2869 /* FIXME Should handle error properly */
2870 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2873 if (crunch_response_triggered(csp, crunchers_light))
2876 * One of the tags created by a server-header
2877 * tagger triggered a crunch. We already
2878 * delivered the crunch response to the client
2879 * and are done here after cleaning up.
2884 /* Buffer and pcrs filter this if appropriate. */
2886 if (!http->ssl) /* We talk plaintext */
2888 content_filter = get_filter_function(csp);
2891 * Only write if we're not buffering for content modification
2893 if (!content_filter)
2895 /* write the server's (modified) header to
2896 * the client (along with anything else that
2897 * may be in the buffer)
2900 if (write_socket(csp->cfd, hdr, strlen(hdr))
2901 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2903 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2905 /* the write failed, so don't bother
2906 * mentioning it to the client...
2907 * it probably can't hear us anyway.
2913 byte_count += (size_t)len;
2916 /* we're finished with the server's header */
2921 /* If this was a MS IIS/5 hack then it means
2922 * the server has already closed the
2923 * connection. Nothing more to read. Time
2928 log_error(LOG_LEVEL_INFO,
2929 "Closed server connection detected with MS IIS5 hack enabled.");
2936 return; /* huh? we should never get here */
2939 if (csp->content_length == 0)
2942 * If Privoxy didn't recalculate the
2943 * Content-Lenght, byte_count is still
2946 csp->content_length = byte_count;
2949 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2950 csp->ip_addr_str, http->ocmd, csp->content_length);
2954 /*********************************************************************
2958 * Description : This is little more than chat. We only "serve" to
2959 * to close any socket that chat may have opened.
2962 * 1 : csp = Current client state (buffers, headers, etc...)
2966 *********************************************************************/
2968 void serve(struct client_state *csp)
2969 #else /* ifndef AMIGA */
2970 static void serve(struct client_state *csp)
2971 #endif /* def AMIGA */
2974 close_socket(csp->cfd);
2976 if (csp->sfd != JB_INVALID_SOCKET)
2978 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2979 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2981 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
2985 forget_connection(csp->sfd);
2986 close_socket(csp->sfd);
2989 close_socket(csp->sfd);
2990 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
2993 csp->flags &= ~CSP_FLAG_ACTIVE;
2999 /*********************************************************************
3001 * Function : server_thread
3003 * Description : We only exist to call `serve' in a threaded environment.
3006 * 1 : data = Current client state (buffers, headers, etc...)
3008 * Returns : Always 0.
3010 *********************************************************************/
3011 static int32 server_thread(void *data)
3013 serve((struct client_state *) data);
3020 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3021 /*********************************************************************
3025 * Description : Print usage info & exit.
3027 * Parameters : Pointer to argv[0] for identifying ourselves
3031 *********************************************************************/
3032 static void usage(const char *myname)
3034 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3038 #endif /* defined(unix) */
3041 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3042 #endif /* defined(unix) */
3043 "[--version] [configfile]\n"
3044 "Aborting\n", myname);
3049 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3052 #ifdef MUTEX_LOCKS_AVAILABLE
3053 /*********************************************************************
3055 * Function : privoxy_mutex_lock
3057 * Description : Locks a mutex.
3060 * 1 : mutex = The mutex to lock.
3062 * Returns : Void. May exit in case of errors.
3064 *********************************************************************/
3065 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3067 #ifdef FEATURE_PTHREAD
3068 int err = pthread_mutex_lock(mutex);
3071 if (mutex != &log_mutex)
3073 log_error(LOG_LEVEL_FATAL,
3074 "Mutex locking failed: %s.\n", strerror(err));
3079 EnterCriticalSection(mutex);
3080 #endif /* def FEATURE_PTHREAD */
3084 /*********************************************************************
3086 * Function : privoxy_mutex_unlock
3088 * Description : Unlocks a mutex.
3091 * 1 : mutex = The mutex to unlock.
3093 * Returns : Void. May exit in case of errors.
3095 *********************************************************************/
3096 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3098 #ifdef FEATURE_PTHREAD
3099 int err = pthread_mutex_unlock(mutex);
3102 if (mutex != &log_mutex)
3104 log_error(LOG_LEVEL_FATAL,
3105 "Mutex unlocking failed: %s.\n", strerror(err));
3110 LeaveCriticalSection(mutex);
3111 #endif /* def FEATURE_PTHREAD */
3115 /*********************************************************************
3117 * Function : privoxy_mutex_init
3119 * Description : Prepares a mutex.
3122 * 1 : mutex = The mutex to initialize.
3124 * Returns : Void. May exit in case of errors.
3126 *********************************************************************/
3127 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3129 #ifdef FEATURE_PTHREAD
3130 int err = pthread_mutex_init(mutex, 0);
3133 printf("Fatal error. Mutex initialization failed: %s.\n",
3138 InitializeCriticalSection(mutex);
3139 #endif /* def FEATURE_PTHREAD */
3141 #endif /* def MUTEX_LOCKS_AVAILABLE */
3143 /*********************************************************************
3145 * Function : initialize_mutexes
3147 * Description : Prepares mutexes if mutex support is available.
3151 * Returns : Void, exits in case of errors.
3153 *********************************************************************/
3154 static void initialize_mutexes(void)
3156 #ifdef MUTEX_LOCKS_AVAILABLE
3158 * Prepare global mutex semaphores
3160 privoxy_mutex_init(&log_mutex);
3161 privoxy_mutex_init(&log_init_mutex);
3162 privoxy_mutex_init(&connection_reuse_mutex);
3165 * XXX: The assumptions below are a bit naive
3166 * and can cause locks that aren't necessary.
3168 * For example older FreeBSD versions (< 6.x?)
3169 * have no gethostbyname_r, but gethostbyname is
3172 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3173 privoxy_mutex_init(&resolver_mutex);
3174 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3176 * XXX: should we use a single mutex for
3177 * localtime() and gmtime() as well?
3179 #ifndef HAVE_GMTIME_R
3180 privoxy_mutex_init(&gmtime_mutex);
3181 #endif /* ndef HAVE_GMTIME_R */
3183 #ifndef HAVE_LOCALTIME_R
3184 privoxy_mutex_init(&localtime_mutex);
3185 #endif /* ndef HAVE_GMTIME_R */
3188 privoxy_mutex_init(&rand_mutex);
3189 #endif /* ndef HAVE_RANDOM */
3190 #endif /* def MUTEX_LOCKS_AVAILABLE */
3194 /*********************************************************************
3198 * Description : Load the config file and start the listen loop.
3199 * This function is a lot more *sane* with the `load_config'
3200 * and `listen_loop' functions; although it stills does
3201 * a *little* too much for my taste.
3204 * 1 : argc = Number of parameters (including $0).
3205 * 2 : argv = Array of (char *)'s to the parameters.
3207 * Returns : 1 if : can't open config file, unrecognized directive,
3208 * stats requested in multi-thread mode, can't open the
3209 * log file, can't open the jar file, listen port is invalid,
3210 * any load fails, and can't bind port.
3212 * Else main never returns, the process must be signaled
3213 * to terminate execution. Or, on Windows, use the
3214 * "File", "Exit" menu option.
3216 *********************************************************************/
3218 int real_main(int argc, const char *argv[])
3220 int main(int argc, const char *argv[])
3224 unsigned int random_seed;
3226 struct passwd *pw = NULL;
3227 struct group *grp = NULL;
3230 char *pre_chroot_nslookup_to_load_resolver = NULL;
3237 #if !defined(_WIN32)
3245 * Parse the command line arguments
3247 * XXX: simply printing usage information in case of
3248 * invalid arguments isn't particularly user friendly.
3250 while (++argc_pos < argc)
3253 /* Check to see if the service must be installed or uninstalled */
3254 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3256 const char *pName = argv[argc_pos] + 9;
3259 exit( (install_service(pName)) ? 0 : 1 );
3261 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3263 const char *pName = argv[argc_pos] + 11;
3266 exit((uninstall_service(pName)) ? 0 : 1);
3268 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3270 bRunAsService = TRUE;
3271 w32_set_service_cwd();
3272 atexit(w32_service_exit_notify);
3275 #endif /* defined(_WIN32) */
3278 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3280 if (strcmp(argv[argc_pos], "--help") == 0)
3285 else if(strcmp(argv[argc_pos], "--version") == 0)
3287 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3293 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3298 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3300 if (++argc_pos == argc) usage(argv[0]);
3301 pidfile = strdup(argv[argc_pos]);
3304 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3306 if (++argc_pos == argc) usage(argv[argc_pos]);
3308 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3311 if (NULL == (grp = getgrnam(p)))
3313 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3317 if (NULL == (pw = getpwnam(argv[argc_pos])))
3319 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3322 if (p != NULL) *--p = '\0';
3325 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3327 if (++argc_pos == argc) usage(argv[0]);
3328 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3331 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3335 #endif /* defined(unix) */
3337 else if (argc_pos + 1 != argc)
3340 * This is neither the last command line
3341 * option, nor was it recognized before,
3342 * therefore it must be invalid.
3348 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3350 configfile = argv[argc_pos];
3353 } /* -END- while (more arguments) */
3356 if ( *configfile != '/' )
3358 char cwd[BUFFER_SIZE];
3360 size_t abs_file_size;
3362 /* make config-filename absolute here */
3363 if (NULL == getcwd(cwd, sizeof(cwd)))
3365 perror("failed to get current working directory");
3370 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3371 basedir = strdup(cwd);
3373 if (NULL == basedir ||
3374 NULL == (abs_file = malloc(abs_file_size)))
3376 perror("malloc failed");
3379 strlcpy(abs_file, basedir, abs_file_size);
3380 strlcat(abs_file, "/", abs_file_size );
3381 strlcat(abs_file, configfile, abs_file_size);
3382 configfile = abs_file;
3384 #endif /* defined unix */
3388 clients->next = NULL;
3390 /* XXX: factor out initialising after the next stable release. */
3393 #elif defined(_WIN32)
3397 /* Prepare mutexes if supported and necessary. */
3398 initialize_mutexes();
3400 /* Enable logging until further notice. */
3401 init_log_module(Argv[0]);
3403 random_seed = (unsigned int)time(NULL);
3405 srandom(random_seed);
3408 #endif /* ifdef HAVE_RANDOM */
3411 * Unix signal handling
3413 * Catch the abort, interrupt and terminate signals for a graceful exit
3414 * Catch the hangup signal so the errlog can be reopened.
3415 * Ignore the broken pipe signals (FIXME: Why?)
3417 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3420 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3421 const int ignored_signals[] = { SIGPIPE, 0 };
3423 for (idx = 0; catched_signals[idx] != 0; idx++)
3425 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3426 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3428 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3429 #endif /* ifdef sun */
3431 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3435 for (idx = 0; ignored_signals[idx] != 0; idx++)
3437 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3439 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3444 #else /* ifdef _WIN32 */
3445 # ifdef _WIN_CONSOLE
3447 * We *are* in a windows console app.
3448 * Print a verbose messages about FAQ's and such
3450 printf("%s", win32_blurb);
3451 # endif /* def _WIN_CONSOLE */
3452 #endif /* def _WIN32 */
3455 /* Initialize the CGI subsystem */
3456 cgi_init_error_messages();
3459 * If runnig on unix and without the --nodaemon
3460 * option, become a daemon. I.e. fork, detach
3461 * from tty and get process group leadership
3474 if ( pid < 0 ) /* error */
3479 else if ( pid != 0 ) /* parent */
3484 * must check for errors
3485 * child died due to missing files aso
3488 wpid = waitpid( pid, &status, WNOHANG );
3497 /* Should be more portable, but not as well tested */
3502 #else /* ndef __FreeBSD__ */
3504 #endif /* ndef __FreeBSD__ */
3505 fd = open("/dev/tty", O_RDONLY);
3508 /* no error check here */
3509 ioctl( fd, TIOCNOTTY,0 );
3514 * stderr (fd 2) will be closed later on, when the
3515 * log file has been parsed.
3522 } /* -END- if (!no_daemon) */
3525 * As soon as we have written the PID file, we can switch
3526 * to the user and group ID indicated by the --user option
3532 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3534 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3540 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3542 /* Read the time zone file from /etc before doing chroot. */
3544 if (NULL != pre_chroot_nslookup_to_load_resolver
3545 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3547 /* Initialize resolver library. */
3548 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3550 if (chroot(pw->pw_dir) < 0)
3552 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3556 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3559 if (setuid(pw->pw_uid))
3561 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3565 char putenv_dummy[64];
3567 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3568 if (putenv(putenv_dummy) != 0)
3570 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3573 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3574 if (putenv(putenv_dummy) != 0)
3576 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3582 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3585 #endif /* defined unix */
3588 /* This will be FALSE unless the command line specified --service
3592 /* Yup, so now we must attempt to establish a connection
3593 * with the service dispatcher. This will only work if this
3594 * process was launched by the service control manager to
3595 * actually run as a service. If this isn't the case, i've
3596 * known it take around 30 seconds or so for the call to return.
3599 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3600 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3602 /* Service has run, and at this point is now being stopped, so just return */
3607 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3609 /* An error occurred. Usually it's because --service was wrongly specified
3610 * and we were unable to connect to the Service Control Dispatcher because
3611 * it wasn't expecting us and is therefore not listening.
3613 * For now, just continue below to call the listen_loop function.
3616 #endif /* def _WIN32 */
3626 /*********************************************************************
3628 * Function : bind_port_helper
3630 * Description : Bind the listen port. Handles logging, and aborts
3634 * 1 : config = Privoxy configuration. Specifies port
3637 * Returns : Port that was opened.
3639 *********************************************************************/
3640 static jb_socket bind_port_helper(struct configuration_spec * config)
3645 if (config->haddr == NULL)
3647 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3652 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3653 config->hport, config->haddr);
3656 result = bind_port(config->haddr, config->hport, &bfd);
3663 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3664 "There may be another Privoxy or some other "
3665 "proxy running on port %d",
3666 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3667 config->hport, config->hport);
3670 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3671 "The hostname is not resolvable",
3672 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3675 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3676 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3679 /* shouldn't get here */
3680 return JB_INVALID_SOCKET;
3683 config->need_bind = 0;
3690 /* Without this simple workaround we get this compiler warning from _beginthread
3691 * warning C4028: formal parameter 1 different from declaration
3693 void w32_service_listen_loop(void *p)
3697 #endif /* def _WIN32 */
3700 /*********************************************************************
3702 * Function : listen_loop
3704 * Description : bind the listen port and enter a "FOREVER" listening loop.
3710 *********************************************************************/
3711 static void listen_loop(void)
3713 struct client_state *csp = NULL;
3715 struct configuration_spec * config;
3717 config = load_config();
3719 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3721 * XXX: Should be relocated once it no
3722 * longer needs to emit log messages.
3724 initialize_reusable_connections();
3725 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3727 bfd = bind_port_helper(config);
3729 #ifdef FEATURE_GRACEFUL_TERMINATION
3730 while (!g_terminate)
3735 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3736 while (waitpid(-1, NULL, WNOHANG) > 0)
3738 /* zombie children */
3740 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3743 * Free data that was used by died threads
3749 * Re-open the errlog after HUP signal
3751 if (received_hup_signal)
3753 init_error_log(Argv[0], config->logfile);
3754 received_hup_signal = 0;
3758 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3760 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3764 csp->flags |= CSP_FLAG_ACTIVE;
3765 csp->sfd = JB_INVALID_SOCKET;
3767 csp->config = config = load_config();
3769 if ( config->need_bind )
3772 * Since we were listening to the "old port", we will not see
3773 * a "listen" param change until the next IJB request. So, at
3774 * least 1 more request must be made for us to find the new
3775 * setting. I am simply closing the old socket and binding the
3778 * Which-ever is correct, we will serve 1 more page via the
3779 * old settings. This should probably be a "show-proxy-args"
3780 * request. This should not be a so common of an operation
3781 * that this will hurt people's feelings.
3786 bfd = bind_port_helper(config);
3789 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3791 if (!accept_connection(csp, bfd))
3793 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3806 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3809 #ifdef FEATURE_TOGGLE
3810 if (global_toggle_state)
3811 #endif /* def FEATURE_TOGGLE */
3813 csp->flags |= CSP_FLAG_TOGGLED_ON;
3816 if (run_loader(csp))
3818 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3819 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3823 if (block_acl(NULL,csp))
3825 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3826 close_socket(csp->cfd);
3830 #endif /* def FEATURE_ACL */
3832 /* add it to the list of clients */
3833 csp->next = clients->next;
3834 clients->next = csp;
3836 if (config->multi_threaded)
3840 /* this is a switch () statment in the C preprocessor - ugh */
3841 #undef SELECTED_ONE_OPTION
3843 /* Use Pthreads in preference to native code */
3844 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3845 #define SELECTED_ONE_OPTION
3847 pthread_t the_thread;
3848 pthread_attr_t attrs;
3850 pthread_attr_init(&attrs);
3851 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3852 errno = pthread_create(&the_thread, &attrs,
3853 (void * (*)(void *))serve, csp);
3854 child_id = errno ? -1 : 0;
3855 pthread_attr_destroy(&attrs);
3859 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3860 #define SELECTED_ONE_OPTION
3861 child_id = _beginthread(
3862 (void (*)(void *))serve,
3867 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3868 #define SELECTED_ONE_OPTION
3869 child_id = _beginthread(
3870 (void(* _Optlink)(void*))serve,
3876 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3877 #define SELECTED_ONE_OPTION
3879 thread_id tid = spawn_thread
3880 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3882 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3884 child_id = (int) tid;
3893 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3894 #define SELECTED_ONE_OPTION
3895 csp->cfd = ReleaseSocket(csp->cfd, -1);
3898 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3899 NP_Output, Output(),
3900 NP_CloseOutput, FALSE,
3901 NP_Name, (ULONG)"privoxy child",
3905 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3906 NP_Output, Output(),
3907 NP_CloseOutput, FALSE,
3908 NP_Name, (ULONG)"privoxy child",
3909 NP_StackSize, 200*1024,
3915 ((struct Task *)child_id)->tc_UserData = csp;
3916 Signal((struct Task *)child_id, SIGF_SINGLE);
3921 #if !defined(SELECTED_ONE_OPTION)
3924 /* This block is only needed when using fork().
3925 * When using threads, the server thread was
3926 * created and run by the call to _beginthread().
3928 if (child_id == 0) /* child */
3931 #ifdef FEATURE_TOGGLE
3932 int inherited_toggle_state = global_toggle_state;
3933 #endif /* def FEATURE_TOGGLE */
3938 * If we've been toggled or we've blocked the request, tell Mom
3941 #ifdef FEATURE_TOGGLE
3942 if (inherited_toggle_state != global_toggle_state)
3944 rc |= RC_FLAG_TOGGLED;
3946 #endif /* def FEATURE_TOGGLE */
3948 #ifdef FEATURE_STATISTICS
3949 if (csp->flags & CSP_FLAG_REJECTED)
3951 rc |= RC_FLAG_BLOCKED;
3953 #endif /* ndef FEATURE_STATISTICS */
3957 else if (child_id > 0) /* parent */
3959 /* in a fork()'d environment, the parent's
3960 * copy of the client socket and the CSP
3964 #if !defined(_WIN32) && !defined(__CYGWIN__)
3966 wait( &child_status );
3969 * Evaluate child's return code: If the child has
3970 * - been toggled, toggle ourselves
3971 * - blocked its request, bump up the stats counter
3974 #ifdef FEATURE_TOGGLE
3975 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3977 global_toggle_state = !global_toggle_state;
3979 #endif /* def FEATURE_TOGGLE */
3981 #ifdef FEATURE_STATISTICS
3983 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3987 #endif /* def FEATURE_STATISTICS */
3989 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3990 close_socket(csp->cfd);
3991 csp->flags &= ~CSP_FLAG_ACTIVE;
3995 #undef SELECTED_ONE_OPTION
3996 /* end of cpp switch () */
3998 if (child_id < 0) /* failed */
4000 char buf[BUFFER_SIZE];
4002 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
4004 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
4006 write_socket(csp->cfd, buf, strlen(buf));
4007 close_socket(csp->cfd);
4008 csp->flags &= ~CSP_FLAG_ACTIVE;
4019 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4021 /* Clean up. Aim: free all memory (no leaks) */
4022 #ifdef FEATURE_GRACEFUL_TERMINATION
4024 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4026 unload_current_config_file();
4027 unload_current_actions_file();
4028 unload_current_re_filterfile();
4029 #ifdef FEATURE_TRUST
4030 unload_current_trust_file();
4033 if (config->multi_threaded)
4040 } while ((clients->next != NULL) && (--i > 0));
4044 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4055 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4056 /* Cleanup - remove taskbar icon etc. */
4061 #endif /* FEATURE_GRACEFUL_TERMINATION */