1 const char jcc_rcs[] = "$Id: jcc.c,v 1.189 2008/10/11 09:53:00 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 1.189 2008/10/11 09:53:00 fabiankeil
37 * Let server_response_is_complete() deal properly with
38 * content that is neither buffered nor read all at once.
40 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
41 * Flush work-in-progress changes to keep outgoing connections
42 * alive where possible. Incomplete and mostly #ifdef'd out.
44 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
45 * Add mutex lock support for _WIN32.
47 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
48 * Prepare for critical sections on Windows by adding a
49 * layer of indirection before the pthread mutex functions.
51 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
52 * Remove FEATURE_COOKIE_JAR.
54 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
55 * - Silence LLVM/Clang complaint.
56 * - Make received_hup_signal static.
57 * - Hide definitions for basedir, pidfile and received_hup_signal
58 * from __EMX__ as they only seem to be used in case of #ifdef unix.
60 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
61 * Accept Shoutcast responses again. Problem reported
62 * and fix suggested by Stefan in #2062860.
64 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
65 * Fix possible NULL-pointer dereference reported
66 * by din_a4 in #2003937. Pointy hat to me.
68 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
69 * Streamline sed()'s prototype and declare
70 * the header parse and add structures static.
72 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
73 * - Mark csp as immutable for send_crunch_response().
74 * - Fix comment spelling.
76 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
77 * Factor update_server_headers() out of sed(), ditch the
78 * first_run hack and make server_patterns_light static.
80 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
81 * Don't provide get_header() with the whole client state
82 * structure when it only needs access to csp->iob.
84 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
85 * Make the "read the rest of the headers" loop a bit more readable.
87 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
88 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
89 * - Remove useless comment.
91 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
92 * Fix comment grammar.
94 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
95 * Remove the pointless buffer in client_protocol_is_unsupported().
97 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
98 * Least-effort fix for bug #1821930 (reported by Lee):
99 * If the response doesn't look like HTTP,
100 * tell the client and log the problem.
102 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
103 * Don't pass the whole csp structure to flush_socket()
104 * when it only needs a file descriptor and a buffer.
106 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
107 * Remove kill-popups action.
109 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
110 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
112 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
113 * Remove the treat-forbidden-connects-like-blocks action. We now
114 * use the "blocked" page for forbidden CONNECT requests by default.
116 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
117 * Also use shiny new connect_port_is_forbidden() in jcc.c.
119 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
120 * Rename url_actions() to get_url_actions() and let it
121 * use the standard parameter ordering.
123 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
124 * Let forward_url() use the standard parameter ordering
125 * and mark its second parameter immutable.
127 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
128 * Remove the "Listening ... for local connections only" log message.
129 * Whether or not remote connections are able to reach Privoxy is up
130 * to the operating system.
132 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
133 * Prevent the log messages for CONNECT requests to unacceptable
134 * ports from printing the limit-connect argument as [null] if
135 * limit-connect hasn't been explicitly enabled.
137 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
138 * Make sure all console-mode apps get a usage() instance
140 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
141 * Reword NO_SERVER_DATA_RESPONSE to make it harder
142 * to misunderstand what the message is all about.
144 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
145 * Unbreak trustfile which previously didn't work without
146 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
148 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
149 * Plug memory leak. Spotted by Valgrind, triggered by
150 * Privoxy-Regression-Test feeding proxyfuzz.py.
152 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
153 * In the HTTP snipplets, refer to the client as client.
155 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
156 * Emit a log message when activating the MS IIS5 hack.
158 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
159 * Log the "weak randomization factor" warning only
160 * once for mingw32 and provide some more details.
162 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
163 * Initialize log module after initializing mutexes, future
164 * deadlocks in that code should now work cross-platform.
166 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
167 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
168 * Reported by Bert van Leeuwen in BR#1818808.
170 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
171 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
173 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
174 * When in daemon mode, close stderr after the configuration file has been
175 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
177 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
178 * - Fix a crash when parsing invalid requests whose first header
179 * is rejected by get_header(). Regression (re?)introduced
180 * in r1.143 by yours truly.
181 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
182 * client_cookie_adder() to make sure send-vanilla-wafer can be
183 * controlled through tags (and thus regression-tested).
185 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
186 * - Move get_filter_function() from jcc.c to filters.c
187 * so the filter functions can be static.
188 * - Don't bother filtering body-less responses.
190 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
191 * Execute content filters through execute_content_filter().
193 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
194 * Initialize req to NULL to make sure it's defined if the
195 * first read_socket() call fails. Reported by icmp30.
197 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
198 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
199 * extensive comments moved to user manual.
201 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
202 * Don't crash if a broken header filter wiped out the request line.
204 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
205 * Fix byte_count calculation in case of flushes
206 * and don't parse the server headers a second time.
208 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
209 * - If there's a connection problem after we already forwarded
210 * parts of the original content, just hang up. Fixes BR#1776724.
211 * - Fix warnings about unused code on mingw32.
212 * - In case of flushes, calculate the byte count
213 * less incorrectly (I think).
215 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
216 * Add some more prototypes for static functions.
218 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
219 * Comment out request_contains_null_bytes() until it's used again.
221 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
222 * #1763173 from Stefan Huehner: s@const static@static const@
223 * and declare some more functions static.
225 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
226 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
227 * and explain why they were rejected in the first place.
228 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
229 * CONNECT requests. The request line was missing.
230 * - Add two more XXX reminders as we don't have enough already.
232 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
233 * As Hal noticed, checking dispatch_cgi() as the last cruncher
234 * looks like a bug if CGI requests are blocked unintentionally,
235 * so don't do it unless the user enabled the new config option
236 * "allow-cgi-request-crunching".
238 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
239 * - Allow to rewrite the request destination behind the client's back.
240 * - Turn the weird-looking unconditional for loop that
241 * reads the client request into a conditional while loop.
242 * Move the stuff that only runs once out of the loop.
243 * - Move parts of chat(), server_content_type() and the
244 * necessary stuff to fix BR#1750917 into get_filter_function().
246 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
247 * Temporary workaround for BR#1730105.
249 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
250 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
251 * deadlocks and crashes on OpenBSD and possibly other OS with neither
252 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
253 * Thanks to Ralf Horstmann for report and solution.
255 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
256 * Add forward-override{} to change the forwarding settings through
257 * action sections. This is mainly interesting to forward different
258 * clients differently (for example based on User-Agent or request
261 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
262 * - Let usage() mention the --chroot parameter.
263 * - Use read_socket() consistently and always leave
264 * the last buffer byte alone, even in cases where
265 * null termination (currently) doesn't matter.
267 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
268 * - Fix config file loading on Unix if no config file is specified.
269 * Since r1.97 Privoxy would always interpret the last argument as
270 * config file, even if it's a valid command line option.
271 * - Abort in case of unrecognized command line options. Closes #1719696.
272 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
273 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
275 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
276 * - Don't rerun crunchers that only depend on the request URL.
277 * - Don't count redirects and CGI requests as "blocked requests".
279 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
280 * Support crunching based on tags created by server-header taggers.
282 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
283 * Make HTTP snippets static (again). Add a Content-Type for those
284 * with content so the browser doesn't guess it based on the URL.
286 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
287 * Move crunching and request line rebuilding out of chat().
289 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
290 * Introduce tags as alternative way to specify which
291 * actions apply to a request. At the moment tags can be
292 * created based on client and server headers.
294 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
295 * Don't CLF-log CONNECT requests twice.
297 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
298 * Log the source address for ACL-related connection drops.
300 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
301 * New config option: enforce-blocks.
303 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
304 * - Move null byte check into separate function.
305 * - Don't confuse the client with error pages
306 * if a CONNECT request was already confirmed.
308 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
309 * Speed up NULL byte escaping and only log the complete
310 * NULL byte requests with header debugging enabled.
312 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
313 * Answer requests that contain NULL bytes with
314 * a custom response instead of waiting for more
315 * data until the client eventually hangs up.
317 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
318 * - Move delivery and logging of crunched responses
319 * from chat() into send_crunch_response().
320 * - Display the reason for generating http_responses.
321 * - Log the content length for LOG_LEVEL_CLF correctly
322 * (still incorrect for some fixed responses).
323 * - Reword an incorrect comment about
324 * treat-forbidden-connects-like-blocks violating
326 * - Add some log messages.
328 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
329 * Move mutex initialization into separate
330 * function and exit in case of errors.
332 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
333 * - Start to reduce chat()'s line count and move
334 * parts of it into separate functions.
335 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
337 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
338 * - Add Proxy-Agent header to HTTP snippets that are
339 * supposed to reach HTTP clients only.
340 * - Made a few CONNECT log messages more descriptive.
341 * - Catch completely empty server responses (as seen
342 * with Tor's fake ".noconnect" top level domain).
343 * - Use shiny new "forwarding-failed" template for socks errors.
345 * Revision 1.118 2007/01/07 07:43:43 joergs
346 * AmigaOS4 support added.
348 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
349 * Added config option accept-intercepted-requests
350 * and disabled it by default.
352 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
353 * Reverted parts of my last commit
354 * to keep error handling working.
356 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
357 * Fixed gcc43 conversion warnings.
359 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
360 * Fix -pedantic ISO C warning about converting
361 * from function pointer to object pointer.
363 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
364 * Silence compiler warning I introduced with my last commit.
366 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
367 * Mutex protect rand() if POSIX threading
368 * is used, warn the user if that's not possible
369 * and stop using it on _WIN32 where it could
372 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
373 * Don't prevent core dumps by catching SIGABRT.
374 * It's rude and makes debugging unreasonable painful.
376 * Revision 1.110 2006/12/13 14:52:53 etresoft
377 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
379 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
380 * Privoxy is now able to run as intercepting
381 * proxy in combination with any packet filter
382 * that does the port redirection. The destination
383 * is extracted from the "Host:" header which
384 * should be available for nearly all requests.
386 * Moved HTTP snipplets into jcc.c.
387 * Added error message for gopher proxy requests.
389 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
390 * Only unlink the pidfile if it's actually used.
392 * Change order of interception checks to make
393 * it possible to block or redirect requests for
396 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
397 * Make pthread mutex locking more generic. Instead of
398 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
399 * and use mutex locking unless there is an _r function
400 * available. Better safe than sorry.
402 * Fixes "./configure --disable-pthread" and should result
403 * in less threading-related problems on pthread-using platforms,
404 * but it still doesn't fix BR#1122404.
406 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
407 * Move pthread.h inclusion from jcc.c to jcc.h.
408 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
410 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
411 * Don't exit after receiving the second SIGHUP on Solaris.
413 * Fixes BR 1052235, but the same problem may exist on other
414 * systems. Once 3.0.6 is out we should use sigset()
415 * where available and see if it breaks anything.
417 * Revision 1.104 2006/09/23 13:26:38 roro
418 * Replace TABs by spaces in source code.
420 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
421 * Fix +redirect{}. Didn't work with -fast-redirects.
423 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
424 * Respond with 400 and a short text message
425 * if the client tries to use Privoxy as FTP proxy.
427 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
428 * Make number of retries in case of forwarded-connect problems
429 * a config file option (forwarded-connect-retries) and use 0 as
432 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
433 * Set random(3) seed.
435 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
436 * Follow the OpenBSD port's lead and protect the resolve
437 * functions on OpenBSD as well.
439 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
440 * --user fix. Only use the user as group if no group is specified.
441 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
443 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
444 * Windows service (re-)integration
448 * --install[:service_name]
449 * --uninstall[:service_name]
452 * They work as follows:
453 * --install will create a service for you and then terminate.
454 * By default the service name will be "privoxy" (without the quotes).
455 * However you can run multiple services if you wish, just by adding
456 * a colon and then a name (no spaces).
458 * --uninstall follows the exact same rules a --install.
460 * --service is used when the program is executed by the service
461 * control manager, and in normal circumstances would never be
462 * used as a command line argument.
464 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
465 * Windows service integration
467 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
468 * Incorporate Fabian Keil's patch work:
469 http://www.fabiankeil.de/sourcecode/privoxy/
471 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
472 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
473 * with what was really the latest development (the v_3_0_branch branch)
475 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
476 * Thanks to Jindrich Makovicka for a race condition fix for the log
477 * file. The race condition remains for non-pthread implementations.
478 * Reference patch #1175720.
480 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
481 * Add the ability to check jpeg images for invalid
482 * lengths of comment blocks. Defensive strategy
483 * against the exploit:
484 * Microsoft Security Bulletin MS04-028
485 * Buffer Overrun in JPEG Processing (GDI+) Could
486 * Allow Code Execution (833987)
487 * Enabled with +inspect-jpegs in actions files.
489 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
490 * - Fixed usage info for non-unix platforms
491 * - Fixed small cmdline parsing bug
493 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
494 * Diagnostics: Now preserve the returncode of pthread_create
495 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
497 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
498 * No longer ignore SIGCHLD. Fixes bug #769381
500 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
501 * Close jarfile on graceful exit, remove stray line
503 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
504 * Cosmetics: Killed a warning, a typo and an allocation left at exit
506 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
507 * No longer rely on non-POSIX.1 extensions of getcwd().
510 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
511 * Replaced setenv() by posix-compliant putenv()
512 * Thanks to Neil McCalden (nmcc AT users.sf.net).
514 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
515 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
517 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
518 * Clean-up and extension of improvements for forked mode:
519 * - Child's return code now consists of flags RC_FLAG_*
520 * - Reporting toggle to parent now properly #ifdef'ed
521 * - Children now report blocking to parent. This enables
522 * statistics in forked mode
524 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
525 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
526 * Children now report having been toggled through _exit(17), parents
527 * watch for that code and toggle themselves if found.
529 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
530 * Wrapping all *_r functions (the non-_r versions of them) with
531 * mutex semaphores for OSX. Hopefully this will take care of all
532 * of those pesky crash reports.
534 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
535 * Fixed two mostly harmless mem leaks
537 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
538 * Fix: Head of global clients list now initialized to NULL
540 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
541 * Added basic support for OPTIONS and TRACE HTTP methods:
542 * - New interceptor direct_response() added in chat().
543 * - sed() moved to earlier in the process, so that the
544 * Host: header is evaluated before actions and forwarding
547 * Revision 1.92 2002/05/08 16:00:46 oes
548 * Chat's buffer handling:
549 * - Fixed bug with unchecked out-of-mem conditions
550 * while reading client request & server headers
551 * - No longer predict if the buffer limit will be exceeded
552 * in the next read -- check add_to_iob's new
553 * return code. If buffer couldn't be extended
554 * (policy or out-of-mem) while
555 * - reading from client: abort
556 * - reading server headers: send error page
557 * - buffering server body for filter: flush,
558 * and if that fails: send error page
560 * Revision 1.91 2002/04/08 20:35:58 swa
563 * Revision 1.90 2002/04/02 14:57:28 oes
564 * Made sending wafers independent of FEATURE_COOKIE_JAR
566 * Revision 1.89 2002/03/31 17:18:59 jongfoster
567 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
569 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
570 * More compiler warning message maintenance
572 * Revision 1.87 2002/03/26 22:29:54 swa
573 * we have a new homepage!
575 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
576 * Workaround for closing the jarfile before load_config() comes around again
578 * Revision 1.85 2002/03/24 15:23:33 jongfoster
581 * Revision 1.84 2002/03/24 13:25:43 swa
582 * name change related issues
584 * Revision 1.83 2002/03/16 23:54:06 jongfoster
585 * Adding graceful termination feature, to help look for memory leaks.
586 * If you enable this (which, by design, has to be done by hand
587 * editing config.h) and then go to http://i.j.b/die, then the program
588 * will exit cleanly after the *next* request. It should free all the
589 * memory that was used.
591 * Revision 1.82 2002/03/13 00:27:05 jongfoster
594 * Revision 1.81 2002/03/12 01:42:50 oes
595 * Introduced modular filters
597 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
598 * OS/2 port maintenance:
599 * - Fixed EMX build - it had decayed a little
600 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
601 * substituted a memset for now.
603 * Revision 1.79 2002/03/09 20:03:52 jongfoster
604 * - Making various functions return int rather than size_t.
605 * (Undoing a recent change). Since size_t is unsigned on
606 * Windows, functions like read_socket that return -1 on
607 * error cannot return a size_t.
609 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
610 * crashes, and also frequently caused JB to jump to 100%
611 * CPU and stay there. (Because it thought it had just
612 * read ((unsigned)-1) == 4Gb of data...)
614 * - The signature of write_socket has changed, it now simply
615 * returns success=0/failure=nonzero.
617 * - Trying to get rid of a few warnings --with-debug on
618 * Windows, I've introduced a new type "jb_socket". This is
619 * used for the socket file descriptors. On Windows, this
620 * is SOCKET (a typedef for unsigned). Everywhere else, it's
621 * an int. The error value can't be -1 any more, so it's
622 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
623 * Windows it maps to the #define INVALID_SOCKET.)
625 * - The signature of bind_port has changed.
627 * Revision 1.78 2002/03/08 21:35:04 oes
628 * Added optional group supplement to --user option. Will now use default group of user if no group given
630 * Revision 1.77 2002/03/07 03:52:06 oes
631 * - Fixed compiler warnings etc
632 * - Improved handling of failed DNS lookups
634 * Revision 1.76 2002/03/06 22:54:35 jongfoster
635 * Automated function-comment nitpicking.
637 * Revision 1.75 2002/03/06 10:02:19 oes
638 * Fixed stupid bug when --user was not given
640 * Revision 1.74 2002/03/06 00:49:31 jongfoster
641 * Fixing warning on Windows
642 * Making #ifdefs that refer to the same variable consistently
643 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
645 * Revision 1.73 2002/03/05 23:57:30 hal9
646 * Stray character 's' on line 1618 was breaking build.
648 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
649 * - Re-enable OS/2 building after new parms were added
650 * - Fix false out of memory report when resolving CGI templates when no IP
651 * address is available of failed attempt (a la no such domain)
653 * Revision 1.71 2002/03/05 18:13:56 oes
654 * Added --user option
656 * Revision 1.70 2002/03/05 04:52:42 oes
657 * Deleted non-errlog debugging code
659 * Revision 1.69 2002/03/04 23:50:00 jongfoster
660 * Splitting off bind_port() call into bind_port_helper(), with
663 * Revision 1.68 2002/03/04 20:17:32 oes
666 * Revision 1.67 2002/03/04 18:18:57 oes
667 * - Removed _DEBUG mode
668 * - Cleand up cmdline parsing
669 * - Introduced --no-daemon, --pidfile options
670 * - Cleaned up signal handling:
671 * - Terminate cleanly on INT, TERM and ABRT
672 * - Schedule logfile for re-opening on HUP
673 * - Ignore CHLD and PIPE
674 * - Leave the rest with their default handlers
675 * - Uniform handler registration
676 * - Added usage() function
677 * - Played styleguide police
679 * Revision 1.66 2002/03/03 15:06:55 oes
680 * Re-enabled automatic config reloading
682 * Revision 1.65 2002/03/03 14:49:11 oes
683 * Fixed CLF logging: Now uses client's original HTTP request
685 * Revision 1.64 2002/03/03 09:18:03 joergs
686 * Made jumbjuster work on AmigaOS again.
688 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
689 * Clean up a little CRLF unpleasantness that suddenly appeared
691 * Revision 1.62 2002/02/20 23:17:23 jongfoster
692 * Detecting some out-of memory conditions and exiting with a log message.
694 * Revision 1.61 2002/01/17 21:01:52 jongfoster
695 * Moving all our URL and URL pattern parsing code to urlmatch.c.
697 * Revision 1.60 2001/12/30 14:07:32 steudten
698 * - Add signal handling (unix)
699 * - Add SIGHUP handler (unix)
700 * - Add creation of pidfile (unix)
701 * - Add action 'top' in rc file (RH)
702 * - Add entry 'SIGNALS' to manpage
703 * - Add exit message to logfile (unix)
705 * Revision 1.59 2001/12/13 14:07:18 oes
706 * Fixed Bug: 503 error page now sent OK
708 * Revision 1.58 2001/11/30 23:37:24 jongfoster
709 * Renaming the Win32 config file to config.txt - this is almost the
710 * same as the corresponding UNIX name "config"
712 * Revision 1.57 2001/11/16 00:47:43 jongfoster
713 * Changing the tty-disconnection code to use setsid().
715 * Revision 1.56 2001/11/13 20:20:54 jongfoster
716 * Tabs->spaces, fixing a bug with missing {} around an if()
718 * Revision 1.55 2001/11/13 20:14:53 jongfoster
719 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
721 * Revision 1.54 2001/11/07 00:03:14 steudten
722 * Give reliable return value if an error
723 * occurs not just 0 with new daemon mode.
725 * Revision 1.53 2001/11/05 21:41:43 steudten
726 * Add changes to be a real daemon just for unix os.
727 * (change cwd to /, detach from controlling tty, set
728 * process group and session leader to the own process.
730 * Add some fatal-error log message for failed malloc().
731 * Add '-d' if compiled with 'configure --with-debug' to
732 * enable debug output.
734 * Revision 1.52 2001/10/26 20:11:20 jongfoster
735 * Fixing type mismatch
737 * Revision 1.51 2001/10/26 17:38:28 oes
740 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
741 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
742 * threads to call select() simultaneously. So, it's time to do a real, live,
743 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
744 * (native). Both versions will work, but using __OS2__ offers multi-threading.
746 * Revision 1.49 2001/10/23 21:41:35 jongfoster
747 * Added call to initialize the (statically-allocated of course)
748 * "out of memory" CGI response.
750 * Revision 1.48 2001/10/10 19:56:46 jongfoster
751 * Moving some code that wasn't cookie-related out of an #ifdef
754 * Revision 1.47 2001/10/10 16:44:36 oes
755 * Added CONNECT destination port limitation check
757 * Revision 1.46 2001/10/08 15:17:41 oes
758 * Re-enabled SSL forwarding
760 * Revision 1.45 2001/10/07 15:42:11 oes
761 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
763 * Moved downgrading of the HTTP version from parse_http_request to
764 * chat(), since we can't decide if it is necessary before we have
765 * determined the actions for the URL. The HTTP command is now
766 * *always* re-built so the repairs need no longer be special-cased.
768 * filter_popups now gets a csp pointer so it can raise the new
769 * CSP_FLAG_MODIFIED flag.
773 * Added configurable size limit for the IOB. If the IOB grows so
774 * large that the next read would exceed the limit, the header
775 * is generated, and the header & unfiltered buffer are flushed
776 * to the client. Chat then continues in non-buffering,
777 * non-filtering body mode.
779 * Revision 1.44 2001/10/02 18:13:57 oes
782 * Revision 1.43 2001/10/02 15:32:13 oes
783 * Moved generation of hdr
785 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
786 * Cleaning up 2 compiler warnings on OS/2.
788 * Revision 1.41 2001/09/16 17:05:14 jongfoster
789 * Removing unused #include showarg.h
791 * Revision 1.40 2001/09/16 15:41:45 jongfoster
792 * Fixing signed/unsigned comparison warning.
794 * Revision 1.39 2001/09/16 13:21:27 jongfoster
795 * Changes to use new list functions.
797 * Revision 1.38 2001/09/16 13:01:46 jongfoster
798 * Removing redundant function call that zeroed zalloc()'d memory.
800 * Revision 1.37 2001/09/10 11:12:24 oes
801 * Deleted unused variable
803 * Revision 1.36 2001/09/10 10:56:15 oes
804 * Silenced compiler warnings
806 * Revision 1.35 2001/07/31 14:44:22 oes
807 * Deleted unused size parameter from filter_popups()
809 * Revision 1.34 2001/07/30 22:08:36 jongfoster
810 * Tidying up #defines:
811 * - All feature #defines are now of the form FEATURE_xxx
812 * - Permanently turned off WIN_GUI_EDIT
813 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
815 * Revision 1.33 2001/07/29 19:32:00 jongfoster
816 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
818 * Revision 1.32 2001/07/29 18:47:05 jongfoster
819 * Adding missing #include "loadcfg.h"
821 * Revision 1.31 2001/07/29 12:17:48 oes
822 * Applied pthread fix by Paul Lieverse
824 * Revision 1.30 2001/07/25 22:57:13 jongfoster
825 * __BEOS__ no longer overrides FEATURE_PTHREAD.
826 * This is because FEATURE_PTHREAD will soon be widely used, so I
827 * want to keep it simple.
829 * Revision 1.29 2001/07/24 12:47:06 oes
830 * Applied BeOS support update by Eugenia
832 * Revision 1.28 2001/07/23 13:26:12 oes
833 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
835 * Revision 1.27 2001/07/19 19:09:47 haroon
836 * - Added code to take care of the situation where while processing the first
837 * server response (which includes the server header), after finding the end
838 * of the headers we were not looking past the end of the headers for
839 * content modification. I enabled it for filter_popups.
840 * Someone else should look to see if other similar operations should be
841 * done to the discarded portion of the buffer.
843 * Note 2001/07/20: No, the other content modification mechanisms will process
844 * the whole iob later anyway. --oes
846 * Revision 1.26 2001/07/18 12:31:36 oes
849 * Revision 1.25 2001/07/15 19:43:49 jongfoster
850 * Supports POSIX threads.
851 * Also removed some unused #includes.
853 * Revision 1.24 2001/07/13 14:00:40 oes
854 * - Generic content modification scheme:
855 * Each feature has its own applicability flag that is set
856 * from csp->action->flags.
857 * Replaced the "filtering" int flag , by a function pointer
858 * "content_filter" to the function that will do the content
859 * modification. If it is != NULL, the document will be buffered
860 * and processed through *content_filter, which must set
861 * csp->content_length and return a modified copy of the body
862 * or return NULL (on failiure).
863 * - Changed csp->is_text to the more generic bitmap csp->content_type
864 * which can currently take the valued CT_TEXT or CT_GIF
866 * - Removed all #ifdef PCRS
868 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
869 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
870 * the `block_acl' call. This prevents a compilation error when the user
871 * does not wish to use the "ACL" feature.
873 * Revision 1.22 2001/06/29 21:45:41 oes
874 * Indentation, CRLF->LF, Tab-> Space
876 * Revision 1.21 2001/06/29 13:29:36 oes
877 * - Cleaned up, improved comments
878 * - Unified all possible interceptors (CGI,
879 * block, trust, fast_redirect) in one
880 * place, with one (CGI) answer generation
881 * mechansim. Much clearer now.
882 * - Removed the GIF image generation, which
883 * is now done in filters.c:block_url()
884 * - Made error conditions like domain lookup
885 * failiure or (various) problems while talking
886 * to the server use cgi.c:error_response()
887 * instead of generating HTML/HTTP in chat() (yuck!)
888 * - Removed logentry from cancelled commit
890 * Revision 1.20 2001/06/09 10:55:28 jongfoster
891 * Changing BUFSIZ ==> BUFFER_SIZE
893 * Revision 1.19 2001/06/07 23:12:52 jongfoster
894 * Replacing function pointer in struct gateway with a directly
895 * called function forwarded_connect().
896 * Replacing struct gateway with struct forward_spec
898 * Revision 1.18 2001/06/03 19:12:16 oes
899 * introduced new cgi handling
901 * Revision 1.17 2001/06/01 20:07:23 jongfoster
902 * Now uses action +image-blocker{} rather than config->tinygif
904 * Revision 1.16 2001/06/01 18:49:17 jongfoster
905 * Replaced "list_share" with "list" - the tiny memory gain was not
906 * worth the extra complexity.
908 * Revision 1.15 2001/05/31 21:24:47 jongfoster
909 * Changed "permission" to "action" throughout.
910 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
911 * Moved vanilla wafer check into chat(), since we must now
912 * decide whether or not to add it based on the URL.
914 * Revision 1.14 2001/05/29 20:14:01 joergs
915 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
918 * Revision 1.13 2001/05/29 09:50:24 jongfoster
919 * Unified blocklist/imagelist/permissionslist.
920 * File format is still under discussion, but the internal changes
923 * Also modified interceptor behaviour:
924 * - We now intercept all URLs beginning with one of the following
925 * prefixes (and *only* these prefixes):
927 * * http://ijbswa.sf.net/config/
928 * * http://ijbswa.sourceforge.net/config/
929 * - New interceptors "home page" - go to http://i.j.b/ to see it.
930 * - Internal changes so that intercepted and fast redirect pages
931 * are not replaced with an image.
932 * - Interceptors now have the option to send a binary page direct
933 * to the client. (i.e. ijb-send-banner uses this)
934 * - Implemented show-url-info interceptor. (Which is why I needed
935 * the above interceptors changes - a typical URL is
936 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
937 * The previous mechanism would not have intercepted that, and
938 * if it had been intercepted then it then it would have replaced
941 * Revision 1.12 2001/05/27 22:17:04 oes
943 * - re_process_buffer no longer writes the modified buffer
944 * to the client, which was very ugly. It now returns the
945 * buffer, which it is then written by chat.
947 * - content_length now adjusts the Content-Length: header
948 * for modified documents rather than crunch()ing it.
949 * (Length info in csp->content_length, which is 0 for
950 * unmodified documents)
952 * - For this to work, sed() is called twice when filtering.
954 * Revision 1.11 2001/05/26 17:27:53 jongfoster
955 * Added support for CLF and fixed LOG_LEVEL_LOG.
956 * Also did CRLF->LF fix of my previous patch.
958 * Revision 1.10 2001/05/26 15:26:15 jongfoster
959 * ACL feature now provides more security by immediately dropping
960 * connections from untrusted hosts.
962 * Revision 1.9 2001/05/26 00:28:36 jongfoster
963 * Automatic reloading of config file.
964 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
965 * Most of the global variables have been moved to a new
966 * struct configuration_spec, accessed through csp->config->globalname
967 * Most of the globals remaining are used by the Win32 GUI.
969 * Revision 1.8 2001/05/25 22:43:18 jongfoster
970 * Fixing minor memory leak and buffer overflow.
972 * Revision 1.7 2001/05/25 22:34:30 jongfoster
975 * Revision 1.6 2001/05/23 00:13:58 joergs
976 * AmigaOS support fixed.
978 * Revision 1.5 2001/05/22 18:46:04 oes
980 * - Enabled filtering banners by size rather than URL
981 * by adding patterns that replace all standard banner
982 * sizes with the "Junkbuster" gif to the re_filterfile
984 * - Enabled filtering WebBugs by providing a pattern
985 * which kills all 1x1 images
987 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
988 * which is selected by the (nonstandard and therefore
989 * capital) letter 'U' in the option string.
990 * It causes the quantifiers to be ungreedy by default.
991 * Appending a ? turns back to greedy (!).
993 * - Added a new interceptor ijb-send-banner, which
994 * sends back the "Junkbuster" gif. Without imagelist or
995 * MSIE detection support, or if tinygif = 1, or the
996 * URL isn't recognized as an imageurl, a lame HTML
997 * explanation is sent instead.
999 * - Added new feature, which permits blocking remote
1000 * script redirects and firing back a local redirect
1002 * The feature is conditionally compiled, i.e. it
1003 * can be disabled with --disable-fast-redirects,
1004 * plus it must be activated by a "fast-redirects"
1005 * line in the config file, has its own log level
1006 * and of course wants to be displayed by show-proxy-args
1007 * Note: Boy, all the #ifdefs in 1001 locations and
1008 * all the fumbling with configure.in and acconfig.h
1009 * were *way* more work than the feature itself :-(
1011 * - Because a generic redirect template was needed for
1012 * this, tinygif = 3 now uses the same.
1014 * - Moved GIFs, and other static HTTP response templates
1017 * - Some minor fixes
1019 * - Removed some >400 CRs again (Jon, you really worked
1022 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1023 * Made failure to bind() a fatal error.
1025 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1026 * Version 2.9.4 checkin.
1027 * - Merged popupfile and cookiefile, and added control over PCRS
1028 * filtering, in new "permissionsfile".
1029 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1030 * file error you now get a message box (in the Win32 GUI) rather
1031 * than the program exiting with no explanation.
1032 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1034 * - Removed tabs from "config"
1035 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1036 * - Bumped up version number.
1038 * Revision 1.2 2001/05/17 22:34:44 oes
1039 * - Added hint on GIF char array generation to jcc.c
1040 * - Cleaned CRLF's from the sources and related files
1041 * - Repaired logging for REF and FRC
1043 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1044 * Initial import of version 2.9.3 source tree
1047 *********************************************************************/
1053 #include <sys/types.h>
1062 # ifndef FEATURE_PTHREAD
1066 # include <windows.h>
1067 # include <process.h>
1068 # endif /* ndef FEATURE_PTHREAD */
1071 # ifndef _WIN_CONSOLE
1072 # include "w32log.h"
1073 # endif /* ndef _WIN_CONSOLE */
1074 # include "w32svrapi.h"
1076 #else /* ifndef _WIN32 */
1078 # if !defined (__OS2__)
1079 # include <unistd.h>
1080 # include <sys/wait.h>
1081 # endif /* ndef __OS2__ */
1082 # include <sys/time.h>
1083 # include <sys/stat.h>
1084 # include <sys/ioctl.h>
1087 #include <sys/termios.h>
1095 # include <signal.h>
1098 # include <socket.h> /* BeOS has select() for sockets only. */
1099 # include <OS.h> /* declarations for threads and stuff. */
1102 # if defined(__EMX__) || defined(__OS2__)
1103 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1108 #define bzero(B,N) memset(B,0x00,n)
1112 # include <select.h>
1117 #include "project.h"
1120 #include "filters.h"
1121 #include "loaders.h"
1122 #include "parsers.h"
1123 #include "miscutil.h"
1125 #include "jbsockets.h"
1126 #include "gateway.h"
1127 #include "actions.h"
1129 #include "loadcfg.h"
1130 #include "urlmatch.h"
1132 const char jcc_h_rcs[] = JCC_H_VERSION;
1133 const char project_h_rcs[] = PROJECT_H_VERSION;
1136 struct client_state clients[1];
1137 struct file_list files[1];
1139 #ifdef FEATURE_STATISTICS
1140 int urls_read = 0; /* total nr of urls read inc rejected */
1141 int urls_rejected = 0; /* total nr of urls rejected */
1142 #endif /* def FEATURE_STATISTICS */
1144 #ifdef FEATURE_GRACEFUL_TERMINATION
1145 int g_terminate = 0;
1148 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1149 static void sig_handler(int the_signal);
1151 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1152 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1153 static jb_err get_server_headers(struct client_state *csp);
1154 static const char *crunch_reason(const struct http_response *rsp);
1155 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1157 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1159 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1160 static jb_err change_request_destination(struct client_state *csp);
1161 static void chat(struct client_state *csp);
1162 static void serve(struct client_state *csp);
1163 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1164 static void usage(const char *myname);
1166 static void initialize_mutexes(void);
1167 static jb_socket bind_port_helper(struct configuration_spec *config);
1168 static void listen_loop(void);
1171 void serve(struct client_state *csp);
1172 #else /* ifndef AMIGA */
1173 static void serve(struct client_state *csp);
1174 #endif /* def AMIGA */
1177 static int32 server_thread(void *data);
1178 #endif /* def __BEOS__ */
1181 #define sleep(N) Sleep(((N) * 1000))
1185 #define sleep(N) DosSleep(((N) * 100))
1188 #ifdef MUTEX_LOCKS_AVAILABLE
1190 * XXX: Does the locking stuff really belong in this file?
1192 privoxy_mutex_t log_mutex;
1193 privoxy_mutex_t log_init_mutex;
1194 privoxy_mutex_t connection_reuse_mutex;
1196 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1197 privoxy_mutex_t resolver_mutex;
1198 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1200 #ifndef HAVE_GMTIME_R
1201 privoxy_mutex_t gmtime_mutex;
1202 #endif /* ndef HAVE_GMTIME_R */
1204 #ifndef HAVE_LOCALTIME_R
1205 privoxy_mutex_t localtime_mutex;
1206 #endif /* ndef HAVE_GMTIME_R */
1209 privoxy_mutex_t rand_mutex;
1210 #endif /* ndef HAVE_RANDOM */
1212 #endif /* def MUTEX_LOCKS_AVAILABLE */
1215 const char *basedir = NULL;
1216 const char *pidfile = NULL;
1217 static int received_hup_signal = 0;
1218 #endif /* defined unix */
1220 /* HTTP snipplets. */
1221 static const char CSUCCEED[] =
1222 "HTTP/1.0 200 Connection established\r\n"
1223 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1225 static const char CHEADER[] =
1226 "HTTP/1.0 400 Invalid header received from client\r\n"
1227 "Proxy-Agent: Privoxy " VERSION "\r\n"
1228 "Content-Type: text/plain\r\n"
1229 "Connection: close\r\n\r\n"
1230 "Invalid header received from client.\r\n";
1232 static const char FTP_RESPONSE[] =
1233 "HTTP/1.0 400 Invalid request received from client\r\n"
1234 "Content-Type: text/plain\r\n"
1235 "Connection: close\r\n\r\n"
1236 "Invalid request. Privoxy doesn't support FTP.\r\n";
1238 static const char GOPHER_RESPONSE[] =
1239 "HTTP/1.0 400 Invalid request received from client\r\n"
1240 "Content-Type: text/plain\r\n"
1241 "Connection: close\r\n\r\n"
1242 "Invalid request. Privoxy doesn't support gopher.\r\n";
1244 /* XXX: should be a template */
1245 static const char MISSING_DESTINATION_RESPONSE[] =
1246 "HTTP/1.0 400 Bad request received from client\r\n"
1247 "Proxy-Agent: Privoxy " VERSION "\r\n"
1248 "Content-Type: text/plain\r\n"
1249 "Connection: close\r\n\r\n"
1250 "Bad request. Privoxy was unable to extract the destination.\r\n";
1252 /* XXX: should be a template */
1253 static const char NO_SERVER_DATA_RESPONSE[] =
1254 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1255 "Proxy-Agent: Privoxy " VERSION "\r\n"
1256 "Content-Type: text/plain\r\n"
1257 "Connection: close\r\n\r\n"
1258 "Empty server or forwarder response.\r\n"
1259 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1261 /* XXX: should be a template */
1262 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1263 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1264 "Proxy-Agent: Privoxy " VERSION "\r\n"
1265 "Content-Type: text/plain\r\n"
1266 "Connection: close\r\n\r\n"
1267 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1270 /* XXX: should be a template */
1271 static const char NULL_BYTE_RESPONSE[] =
1272 "HTTP/1.0 400 Bad request received from client\r\n"
1273 "Proxy-Agent: Privoxy " VERSION "\r\n"
1274 "Content-Type: text/plain\r\n"
1275 "Connection: close\r\n\r\n"
1276 "Bad request. Null byte(s) before end of request.\r\n";
1279 /* XXX: should be a template */
1280 static const char MESSED_UP_REQUEST_RESPONSE[] =
1281 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1282 "Proxy-Agent: Privoxy " VERSION "\r\n"
1283 "Content-Type: text/plain\r\n"
1284 "Connection: close\r\n\r\n"
1285 "Bad request. Messed up with header filters.\r\n";
1287 /* A function to crunch a response */
1288 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1290 /* Crunch function flags */
1291 #define CF_NO_FLAGS 0
1292 /* Cruncher applies to forced requests as well */
1293 #define CF_IGNORE_FORCE 1
1294 /* Crunched requests are counted for the block statistics */
1295 #define CF_COUNT_AS_REJECT 2
1297 /* A crunch function and its flags */
1300 const crunch_func_ptr cruncher;
1304 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1306 /* Complete list of cruncher functions */
1307 static const struct cruncher crunchers_all[] = {
1308 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1309 { block_url, CF_COUNT_AS_REJECT },
1310 #ifdef FEATURE_TRUST
1311 { trust_url, CF_COUNT_AS_REJECT },
1312 #endif /* def FEATURE_TRUST */
1313 { redirect_url, CF_NO_FLAGS },
1314 { dispatch_cgi, CF_IGNORE_FORCE},
1318 /* Light version, used after tags are applied */
1319 static const struct cruncher crunchers_light[] = {
1320 { block_url, CF_COUNT_AS_REJECT },
1321 { redirect_url, CF_NO_FLAGS },
1327 * XXX: Don't we really mean
1333 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1334 /*********************************************************************
1336 * Function : sig_handler
1338 * Description : Signal handler for different signals.
1339 * Exit gracefully on TERM and INT
1340 * or set a flag that will cause the errlog
1341 * to be reopened by the main thread on HUP.
1344 * 1 : the_signal = the signal cause this function to call
1348 *********************************************************************/
1349 static void sig_handler(int the_signal)
1355 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1367 received_hup_signal = 1;
1373 * We shouldn't be here, unless we catch signals
1374 * in main() that we can't handle here!
1376 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1384 /*********************************************************************
1386 * Function : client_protocol_is_unsupported
1388 * Description : Checks if the client used a known unsupported
1389 * protocol and deals with it by sending an error
1393 * 1 : csp = Current client state (buffers, headers, etc...)
1394 * 2 : req = the first request line send by the client
1396 * Returns : TRUE if an error response has been generated, or
1397 * FALSE if the request doesn't look invalid.
1399 *********************************************************************/
1400 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1403 * If it's a FTP or gopher request, we don't support it.
1405 * These checks are better than nothing, but they might
1406 * not work in all configurations and some clients might
1407 * have problems digesting the answer.
1409 * They should, however, never cause more problems than
1410 * Privoxy's old behaviour (returning the misleading HTML
1413 * "Could not resolve http://(ftp|gopher)://example.org").
1415 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1417 const char *response = NULL;
1418 const char *protocol = NULL;
1420 if (!strncmpic(req, "GET ftp://", 10))
1422 response = FTP_RESPONSE;
1427 response = GOPHER_RESPONSE;
1428 protocol = "GOPHER";
1430 log_error(LOG_LEVEL_ERROR,
1431 "%s tried to use Privoxy as %s proxy: %s",
1432 csp->ip_addr_str, protocol, req);
1433 log_error(LOG_LEVEL_CLF,
1434 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1436 write_socket(csp->cfd, response, strlen(response));
1445 /*********************************************************************
1447 * Function : get_request_destination_elsewhere
1449 * Description : If the client's request was redirected into
1450 * Privoxy without the client's knowledge,
1451 * the request line lacks the destination host.
1453 * This function tries to get it elsewhere,
1454 * provided accept-intercepted-requests is enabled.
1456 * "Elsewhere" currently only means "Host: header",
1457 * but in the future we may ask the redirecting
1458 * packet filter to look the destination up.
1460 * If the destination stays unknown, an error
1461 * response is send to the client and headers
1462 * are freed so that chat() can return directly.
1465 * 1 : csp = Current client state (buffers, headers, etc...)
1466 * 2 : headers = a header list
1468 * Returns : JB_ERR_OK if the destination is now known, or
1469 * JB_ERR_PARSE if it isn't.
1471 *********************************************************************/
1472 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1476 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1478 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1479 " Privoxy isn't configured to accept intercepted requests.",
1480 csp->ip_addr_str, csp->http->cmd);
1481 /* XXX: Use correct size */
1482 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1483 csp->ip_addr_str, csp->http->cmd);
1485 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1486 destroy_list(headers);
1488 return JB_ERR_PARSE;
1490 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1492 /* Split the domain we just got for pattern matching */
1493 init_domain_components(csp->http);
1499 /* We can't work without destination. Go spread the news.*/
1501 req = list_to_text(headers);
1503 /* XXX: Use correct size */
1504 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1505 csp->ip_addr_str, csp->http->cmd);
1506 log_error(LOG_LEVEL_ERROR,
1507 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1508 csp->ip_addr_str, csp->http->cmd, req);
1511 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1512 destroy_list(headers);
1514 return JB_ERR_PARSE;
1517 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1518 * to get the destination IP address, use it as host directly
1519 * or do a reverse DNS lookup first.
1524 /*********************************************************************
1526 * Function : get_server_headers
1528 * Description : Parses server headers in iob and fills them
1529 * into csp->headers so that they can later be
1533 * 1 : csp = Current client state (buffers, headers, etc...)
1535 * Returns : JB_ERR_OK if everything went fine, or
1536 * JB_ERR_PARSE if the headers were incomplete.
1538 *********************************************************************/
1539 static jb_err get_server_headers(struct client_state *csp)
1541 int continue_hack_in_da_house = 0;
1544 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1549 * continue hack in da house. Ignore the ending of
1550 * this head and continue enlisting header lines.
1551 * The reason is described below.
1553 enlist(csp->headers, "");
1554 continue_hack_in_da_house = 0;
1557 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1560 * It's a bodyless continue response, don't
1561 * stop header parsing after reaching its end.
1563 * As a result Privoxy will concatenate the
1564 * next response's head and parse and deliver
1565 * the headers as if they belonged to one request.
1567 * The client will separate them because of the
1568 * empty line between them.
1570 * XXX: What we're doing here is clearly against
1571 * the intended purpose of the continue header,
1572 * and under some conditions (HTTP/1.0 client request)
1573 * it's a standard violation.
1575 * Anyway, "sort of against the spec" is preferable
1576 * to "always getting confused by Continue responses"
1577 * (Privoxy's behaviour before this hack was added)
1579 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1580 continue_hack_in_da_house = 1;
1582 else if (*header == '\0')
1585 * If the header is empty, but the Continue hack
1586 * isn't active, we can assume that we reached the
1587 * end of the buffer before we hit the end of the
1590 * Inform the caller an let it decide how to handle it.
1592 return JB_ERR_PARSE;
1595 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1598 * XXX: Should we quit the request and return a
1599 * out of memory error page instead?
1601 log_error(LOG_LEVEL_ERROR,
1602 "Out of memory while enlisting server headers. %s lost.",
1612 /*********************************************************************
1614 * Function : crunch_reason
1616 * Description : Translates the crunch reason code into a string.
1619 * 1 : rsp = a http_response
1621 * Returns : A string with the crunch reason or an error description.
1623 *********************************************************************/
1624 static const char *crunch_reason(const struct http_response *rsp)
1626 char * reason = NULL;
1628 assert(rsp != NULL);
1631 return "Internal error while searching for crunch reason";
1634 switch (rsp->reason)
1636 case RSP_REASON_UNSUPPORTED:
1637 reason = "Unsupported HTTP feature";
1639 case RSP_REASON_BLOCKED:
1642 case RSP_REASON_UNTRUSTED:
1643 reason = "Untrusted";
1645 case RSP_REASON_REDIRECTED:
1646 reason = "Redirected";
1648 case RSP_REASON_CGI_CALL:
1649 reason = "CGI Call";
1651 case RSP_REASON_NO_SUCH_DOMAIN:
1652 reason = "DNS failure";
1654 case RSP_REASON_FORWARDING_FAILED:
1655 reason = "Forwarding failed";
1657 case RSP_REASON_CONNECT_FAILED:
1658 reason = "Connection failure";
1660 case RSP_REASON_OUT_OF_MEMORY:
1661 reason = "Out of memory (may mask other reasons)";
1664 reason = "No reason recorded";
1672 /*********************************************************************
1674 * Function : send_crunch_response
1676 * Description : Delivers already prepared response for
1677 * intercepted requests, logs the interception
1678 * and frees the response.
1681 * 1 : csp = Current client state (buffers, headers, etc...)
1682 * 1 : rsp = Fully prepared response. Will be freed on exit.
1684 * Returns : Nothing.
1686 *********************************************************************/
1687 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1689 const struct http_request *http = csp->http;
1690 char status_code[4];
1692 assert(rsp != NULL);
1693 assert(rsp->head != NULL);
1698 * Not supposed to happen. If it does
1699 * anyway, treat it as an unknown error.
1701 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1702 /* return code doesn't matter */
1707 /* If rsp is still NULL, we have serious internal problems. */
1708 log_error(LOG_LEVEL_FATAL,
1709 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1713 * Extract the status code from the actual head
1714 * that was send to the client. It is the only
1715 * way to get it right for all requests, including
1716 * the fixed ones for out-of-memory problems.
1718 * A head starts like this: 'HTTP/1.1 200...'
1722 status_code[0] = rsp->head[9];
1723 status_code[1] = rsp->head[10];
1724 status_code[2] = rsp->head[11];
1725 status_code[3] = '\0';
1727 /* Write the answer to the client */
1728 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1729 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1731 /* There is nothing we can do about it. */
1732 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1735 /* Log that the request was crunched and why. */
1736 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1737 http->hostport, http->path, crunch_reason(rsp));
1738 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1739 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1741 /* Clean up and return */
1742 if (cgi_error_memory() != rsp)
1744 free_http_response(rsp);
1751 /*********************************************************************
1753 * Function : request_contains_null_bytes
1755 * Description : Checks for NULL bytes in the request and sends
1756 * an error message to the client if any were found.
1758 * XXX: currently not used, see comment in chat().
1761 * 1 : csp = Current client state (buffers, headers, etc...)
1762 * 2 : buf = Data from the client's request to check.
1763 * 3 : len = The data length.
1765 * Returns : TRUE if the request contained one or more NULL bytes, or
1768 *********************************************************************/
1769 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1771 size_t c_len; /* Request lenght when treated as C string */
1773 c_len = strlen(buf);
1778 * Null byte(s) found. Log the request,
1779 * return an error response and hang up.
1781 size_t tmp_len = c_len;
1786 * Replace NULL byte(s) with '°' characters
1787 * so the request can be logged as string.
1788 * XXX: Is there a better replacement character?
1791 tmp_len += strlen(buf+tmp_len);
1792 } while (tmp_len < len);
1794 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1795 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1796 log_error(LOG_LEVEL_HEADER,
1797 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1799 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1801 /* XXX: Log correct size */
1802 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1812 /*********************************************************************
1814 * Function : crunch_response_triggered
1816 * Description : Checks if the request has to be crunched,
1817 * and delivers the crunch response if necessary.
1820 * 1 : csp = Current client state (buffers, headers, etc...)
1821 * 2 : crunchers = list of cruncher functions to run
1823 * Returns : TRUE if the request was answered with a crunch response
1826 *********************************************************************/
1827 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1829 struct http_response *rsp = NULL;
1830 const struct cruncher *c;
1833 * If CGI request crunching is disabled,
1834 * check the CGI dispatcher out of order to
1835 * prevent unintentional blocks or redirects.
1837 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1838 && (NULL != (rsp = dispatch_cgi(csp))))
1840 /* Deliver, log and free the interception response. */
1841 send_crunch_response(csp, rsp);
1845 for (c = crunchers; c->cruncher != NULL; c++)
1848 * Check the cruncher if either Privoxy is toggled
1849 * on and the request isn't forced, or if the cruncher
1850 * applies to forced requests as well.
1852 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1853 !(csp->flags & CSP_FLAG_FORCED)) ||
1854 (c->flags & CF_IGNORE_FORCE))
1856 rsp = c->cruncher(csp);
1859 /* Deliver, log and free the interception response. */
1860 send_crunch_response(csp, rsp);
1861 #ifdef FEATURE_STATISTICS
1862 if (c->flags & CF_COUNT_AS_REJECT)
1864 csp->flags |= CSP_FLAG_REJECTED;
1866 #endif /* def FEATURE_STATISTICS */
1877 /*********************************************************************
1879 * Function : build_request_line
1881 * Description : Builds the HTTP request line.
1883 * If a HTTP forwarder is used it expects the whole URL,
1884 * web servers only get the path.
1887 * 1 : csp = Current client state (buffers, headers, etc...)
1888 * 2 : fwd = The forwarding spec used for the request
1889 * XXX: Should use http->fwd instead.
1890 * 3 : request_line = The old request line which will be replaced.
1892 * Returns : Nothing. Terminates in case of memory problems.
1894 *********************************************************************/
1895 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1897 struct http_request *http = csp->http;
1899 assert(http->ssl == 0);
1902 * Downgrade http version from 1.1 to 1.0
1903 * if +downgrade action applies.
1905 if ( (csp->action->flags & ACTION_DOWNGRADE)
1906 && (!strcmpic(http->ver, "HTTP/1.1")))
1909 http->ver = strdup("HTTP/1.0");
1911 if (http->ver == NULL)
1913 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1918 * Rebuild the request line.
1920 freez(*request_line);
1921 *request_line = strdup(http->gpc);
1922 string_append(request_line, " ");
1924 if (fwd->forward_host)
1926 string_append(request_line, http->url);
1930 string_append(request_line, http->path);
1932 string_append(request_line, " ");
1933 string_append(request_line, http->ver);
1935 if (*request_line == NULL)
1937 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1939 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1943 /*********************************************************************
1945 * Function : change_request_destination
1947 * Description : Parse a (rewritten) request line and regenerate
1948 * the http request data.
1951 * 1 : csp = Current client state (buffers, headers, etc...)
1953 * Returns : Forwards the parse_http_request() return code.
1954 * Terminates in case of memory problems.
1956 *********************************************************************/
1957 static jb_err change_request_destination(struct client_state *csp)
1959 struct http_request *http = csp->http;
1962 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1963 free_http_request(http);
1964 err = parse_http_request(csp->headers->first->str, http, csp);
1965 if (JB_ERR_OK != err)
1967 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1968 jb_err_to_string(err));
1972 /* XXX: ocmd is a misleading name */
1973 http->ocmd = strdup(http->cmd);
1974 if (http->ocmd == NULL)
1976 log_error(LOG_LEVEL_FATAL,
1977 "Out of memory copying rewritten HTTP request line");
1985 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
1986 /*********************************************************************
1988 * Function : server_response_is_complete
1990 * Description : Determines whether we should stop reading
1991 * from the server socket.
1994 * 1 : csp = Current client state (buffers, headers, etc...)
1995 * 2 : content_length = Length of content received so far.
1997 * Returns : TRUE if the response is complete,
2000 *********************************************************************/
2001 static int server_response_is_complete(struct client_state *csp, size_t content_length)
2003 int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2005 if (!strcmpic(csp->http->gpc, "HEAD"))
2008 * "HEAD" implies no body, we are thus expecting
2009 * no content. XXX: incomplete "list" of methods?
2011 log_error(LOG_LEVEL_INFO, "Method %s implies no body.", csp->http->gpc);
2012 csp->expected_content_length = 0;
2013 content_length_known = TRUE;
2016 if (csp->http->status == 304)
2019 * Expect no body. XXX: incomplete "list" of status codes?
2021 log_error(LOG_LEVEL_INFO, "Status code %d implies no body.", csp->http->status);
2022 csp->expected_content_length = 0;
2023 content_length_known = TRUE;
2026 return (content_length_known && ((0 == csp->expected_content_length)
2027 || (csp->expected_content_length <= content_length)));
2029 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2032 /*********************************************************************
2036 * Description : Once a connection to the client has been accepted,
2037 * this function is called (via serve()) to handle the
2038 * main business of the communication. When this
2039 * function returns, the caller must close the client
2042 * FIXME: chat is nearly thousand lines long.
2046 * 1 : csp = Current client state (buffers, headers, etc...)
2048 * Returns : Nothing.
2050 *********************************************************************/
2051 static void chat(struct client_state *csp)
2053 char buf[BUFFER_SIZE];
2061 int ms_iis5_hack = 0;
2062 size_t byte_count = 0;
2063 int forwarded_connect_retries = 0;
2064 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2065 const struct forward_spec * fwd;
2066 struct http_request *http;
2067 int len; /* for buffer sizes (and negative error codes) */
2070 /* Function that does the content filtering for the current request */
2071 filter_function_ptr content_filter = NULL;
2073 /* Skeleton for HTTP response, if we should intercept the request */
2074 struct http_response *rsp;
2076 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2077 struct list header_list;
2078 struct list *headers = &header_list;
2082 memset(buf, 0, sizeof(buf));
2085 * Read the client's request. Note that since we're not using select() we
2086 * could get blocked here if a client connected, then didn't say anything!
2091 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2093 if (len <= 0) break; /* error! */
2096 * If there is no memory left for buffering the
2097 * request, there is nothing we can do but hang up
2099 if (add_to_iob(csp, buf, len))
2104 req = get_header(csp->iob);
2106 } while ((NULL != req) && ('\0' == *req));
2108 if ((NULL != req) && ('\0' != *req))
2110 /* Request received. Validate and parse it. */
2114 * XXX: Temporary disabled to prevent problems
2115 * with POST requests whose bodies are allowed to
2116 * contain NULL bytes. BR#1730105.
2118 * The main purpose of this check is to properly
2119 * log stuff like BitTorrent traffic and other junk
2120 * that hits public proxies. It's not required for
2121 * Privoxy to functions as those requests are discarded
2124 * It probably should be rewritten to only check
2125 * the head of the request. Another option would
2126 * be to let all POST requests pass, although that
2127 * may not be good enough.
2129 if (request_contains_null_bytes(csp, buf, len))
2131 /* NULL bytes found and dealt with, just hang up. */
2136 /* Does the request line look invalid? */
2137 if (client_protocol_is_unsupported(csp, req))
2140 * Yes. The request has already been
2141 * answered with a error response, the buffers
2142 * were freed and we're done with chatting.
2147 #ifdef FEATURE_FORCE_LOAD
2149 * If this request contains the FORCE_PREFIX and blocks
2150 * aren't enforced, get rid of it and set the force flag.
2152 if (strstr(req, FORCE_PREFIX))
2154 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2156 log_error(LOG_LEVEL_FORCE,
2157 "Ignored force prefix in request: \"%s\".", req);
2161 strclean(req, FORCE_PREFIX);
2162 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2163 csp->flags |= CSP_FLAG_FORCED;
2167 #endif /* def FEATURE_FORCE_LOAD */
2168 err = parse_http_request(req, http, csp);
2169 if (JB_ERR_OK != err)
2171 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2177 if (http->cmd == NULL)
2179 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2180 /* XXX: Use correct size */
2181 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2182 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2184 free_http_request(http);
2188 /* grab the rest of the client's headers */
2192 p = get_header(csp->iob);
2196 /* There are no additional headers to read. */
2203 * We didn't receive a complete header
2204 * line yet, get the rest of it.
2206 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2209 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2210 destroy_list(headers);
2214 if (add_to_iob(csp, buf, len))
2217 * If there is no memory left for buffering the
2218 * request, there is nothing we can do but hang up
2220 destroy_list(headers);
2227 * We were able to read a complete
2228 * header and can finaly enlist it.
2235 if (http->host == NULL)
2238 * If we still don't know the request destination,
2239 * the request is invalid or the client uses
2240 * Privoxy without its knowledge.
2242 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2245 * Our attempts to get the request destination
2246 * elsewhere failed or Privoxy is configured
2247 * to only accept proxy requests.
2249 * An error response has already been send
2250 * and we're done here.
2257 * Determine the actions for this URL
2259 #ifdef FEATURE_TOGGLE
2260 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2262 /* Most compatible set of actions (i.e. none) */
2263 init_current_action(csp->action);
2266 #endif /* ndef FEATURE_TOGGLE */
2268 get_url_actions(csp, http);
2272 * Save a copy of the original request for logging
2274 http->ocmd = strdup(http->cmd);
2276 if (http->ocmd == NULL)
2278 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2281 enlist(csp->headers, http->cmd);
2283 /* Append the previously read headers */
2284 list_append_list_unique(csp->headers, headers);
2285 destroy_list(headers);
2287 err = sed(csp, FILTER_CLIENT_HEADERS);
2288 if (JB_ERR_OK != err)
2290 assert(err == JB_ERR_PARSE);
2291 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2293 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2295 /* Check request line for rewrites. */
2296 if ((NULL == csp->headers->first->str)
2297 || (strcmp(http->cmd, csp->headers->first->str) &&
2298 (JB_ERR_OK != change_request_destination(csp))))
2301 * A header filter broke the request line - bail out.
2303 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2304 /* XXX: Use correct size */
2305 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2306 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2308 free_http_request(http);
2312 /* decide how to route the HTTP request */
2313 fwd = forward_url(csp, http);
2316 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2317 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2321 * build the http request to send to the server
2322 * we have to do one of the following:
2324 * create = use the original HTTP request to create a new
2325 * HTTP request that has either the path component
2326 * without the http://domainspec (w/path) or the
2327 * full orininal URL (w/url)
2328 * Note that the path and/or the HTTP version may
2329 * have been altered by now.
2331 * connect = Open a socket to the host:port of the server
2332 * and short-circuit server and client socket.
2334 * pass = Pass the request unchanged if forwarding a CONNECT
2335 * request to a parent proxy. Note that we'll be sending
2336 * the CFAIL message ourselves if connecting to the parent
2337 * fails, but we won't send a CSUCCEED message if it works,
2338 * since that would result in a double message (ours and the
2339 * parent's). After sending the request to the parent, we simply
2342 * here's the matrix:
2345 * +--------+--------+
2347 * 0 | create | connect|
2349 * Forwarding +--------+--------+
2351 * 1 | create | pass |
2353 * +--------+--------+
2357 if (http->ssl && connect_port_is_forbidden(csp))
2359 const char *acceptable_connect_ports =
2360 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2361 assert(NULL != acceptable_connect_ports);
2362 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2363 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2364 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2365 csp->action->flags |= ACTION_BLOCK;
2371 freez(csp->headers->first->str);
2372 build_request_line(csp, fwd, &csp->headers->first->str);
2375 hdr = list_to_text(csp->headers);
2378 /* FIXME Should handle error properly */
2379 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2383 * We have a request. Check if one of the crunchers wants it.
2385 if (crunch_response_triggered(csp, crunchers_all))
2388 * Yes. The client got the crunch response
2389 * and we are done here after cleaning up.
2392 list_remove_all(csp->headers);
2398 * The headers can't be removed earlier because
2399 * they were still needed for the referrer check
2400 * in case of CGI crunches.
2402 * XXX: Would it be worth to move the referrer check
2403 * into client_referrer() and set a flag if it's trusted?
2405 list_remove_all(csp->headers);
2407 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2409 if (fwd->forward_host)
2411 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2412 fwd->forward_host, fwd->forward_port, http->hostport);
2416 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2419 /* here we connect to the server, gateway, or the forwarder */
2421 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2422 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2424 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2425 forwarded_connect_retries, http->hostport);
2428 if (csp->sfd == JB_INVALID_SOCKET)
2430 if (fwd->type != SOCKS_NONE)
2433 rsp = error_response(csp, "forwarding-failed", errno);
2435 else if (errno == EINVAL)
2437 rsp = error_response(csp, "no-such-domain", errno);
2441 rsp = error_response(csp, "connect-failed", errno);
2442 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2447 /* Write the answer to the client */
2450 send_crunch_response(csp, rsp);
2457 if (fwd->forward_host || (http->ssl == 0))
2459 /* write the client's (modified) header to the server
2460 * (along with anything else that may be in the buffer)
2463 if (write_socket(csp->sfd, hdr, strlen(hdr))
2464 || (flush_socket(csp->sfd, csp->iob) < 0))
2466 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2469 rsp = error_response(csp, "connect-failed", errno);
2473 send_crunch_response(csp, rsp);
2483 * We're running an SSL tunnel and we're not forwarding,
2484 * so just send the "connect succeeded" message to the
2485 * client, flush the rest, and get out of the way.
2487 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2495 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2497 /* we're finished with the client's header */
2500 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2502 /* pass data between the client and server
2503 * until one or the other shuts down the connection.
2512 * FD_ZERO here seems to point to an errant macro which crashes.
2513 * So do this by hand for now...
2515 memset(&rfds,0x00,sizeof(fd_set));
2519 FD_SET(csp->cfd, &rfds);
2520 FD_SET(csp->sfd, &rfds);
2522 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2523 if (server_body && server_response_is_complete(csp, byte_count))
2525 log_error(LOG_LEVEL_CONNECT,
2526 "Done reading from server. Expected content length: %d. "
2527 "Actual content length: %d. Most recently received: %d.",
2528 csp->expected_content_length, byte_count, len);
2531 * XXX: should not jump around,
2532 * chat() is complicated enough already.
2536 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2538 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2542 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2546 /* this is the body of the browser's request
2547 * just read it and write it.
2550 if (FD_ISSET(csp->cfd, &rfds))
2552 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2556 break; /* "game over, man" */
2559 if (write_socket(csp->sfd, buf, (size_t)len))
2561 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2568 * The server wants to talk. It could be the header or the body.
2569 * If `hdr' is null, then it's the header otherwise it's the body.
2570 * FIXME: Does `hdr' really mean `host'? No.
2574 if (FD_ISSET(csp->sfd, &rfds))
2577 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2581 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2583 if (http->ssl && (fwd->forward_host == NULL))
2586 * Just hang up. We already confirmed the client's CONNECT
2587 * request with status code 200 and unencrypted content is
2588 * no longer welcome.
2590 log_error(LOG_LEVEL_ERROR,
2591 "CONNECT already confirmed. Unable to tell the client about the problem.");
2594 else if (byte_count)
2597 * Just hang up. We already transmitted the original headers
2598 * and parts of the original content and therefore missed the
2599 * chance to send an error message (without risking data corruption).
2601 * XXX: we could retry with a fancy range request here.
2603 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2604 "Unable to tell the client about the problem.");
2608 rsp = error_response(csp, "connect-failed", errno);
2612 send_crunch_response(csp, rsp);
2618 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2619 if (csp->flags & CSP_FLAG_CHUNKED)
2621 if ((len > 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
2623 /* XXX: this is a temporary hack */
2624 log_error(LOG_LEVEL_CONNECT,
2625 "Looks like we reached the end of the last chunk: "
2626 "%d %d %d %d %d. We better stop reading.",
2627 buf[len-5], buf[len-4], buf[len-3], buf[len-2], buf[len-1]);
2628 csp->expected_content_length = byte_count + len;
2629 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2633 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2635 /* Add a trailing zero. This lets filter_popups
2636 * use string operations.
2640 /* Normally, this would indicate that we've read
2641 * as much as the server has sent us and we can
2642 * close the client connection. However, Microsoft
2643 * in its wisdom has released IIS/5 with a bug that
2644 * prevents it from sending the trailing \r\n in
2645 * a 302 redirect header (and possibly other headers).
2646 * To work around this if we've haven't parsed
2647 * a full header we'll append a trailing \r\n
2648 * and see if this now generates a valid one.
2650 * This hack shouldn't have any impacts. If we've
2651 * already transmitted the header or if this is a
2652 * SSL connection, then we won't bother with this
2653 * hack. So we only work on partially received
2654 * headers. If we append a \r\n and this still
2655 * doesn't generate a valid header, then we won't
2656 * transmit anything to the client.
2661 if (server_body || http->ssl)
2664 * If we have been buffering up the document,
2665 * now is the time to apply content modification
2666 * and send the result to the client.
2670 p = execute_content_filter(csp, content_filter);
2672 * If the content filter fails, use the original
2673 * buffer and length.
2674 * (see p != NULL ? p : csp->iob->cur below)
2678 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2681 if (JB_ERR_OK != update_server_headers(csp))
2683 log_error(LOG_LEVEL_FATAL,
2684 "Failed to update server headers. after filtering.");
2687 hdr = list_to_text(csp->headers);
2690 /* FIXME Should handle error properly */
2691 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2694 if (write_socket(csp->cfd, hdr, strlen(hdr))
2695 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2697 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2707 break; /* "game over, man" */
2711 * This is NOT the body, so
2712 * Let's pretend the server just sent us a blank line.
2714 snprintf(buf, sizeof(buf), "\r\n");
2715 len = (int)strlen(buf);
2718 * Now, let the normal header parsing algorithm below do its
2719 * job. If it fails, we'll exit instead of continuing.
2726 * If this is an SSL connection or we're in the body
2727 * of the server document, just write it to the client,
2728 * unless we need to buffer the body for later content-filtering
2731 if (server_body || http->ssl)
2736 * If there is no memory left for buffering the content, or the buffer limit
2737 * has been reached, switch to non-filtering mode, i.e. make & write the
2738 * header, flush the iob and buf, and get out of the way.
2740 if (add_to_iob(csp, buf, len))
2745 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2747 hdr = list_to_text(csp->headers);
2751 * Memory is too tight to even generate the header.
2752 * Send our static "Out-of-memory" page.
2754 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2755 rsp = cgi_error_memory();
2756 send_crunch_response(csp, rsp);
2760 hdrlen = strlen(hdr);
2762 if (write_socket(csp->cfd, hdr, hdrlen)
2763 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2764 || (write_socket(csp->cfd, buf, (size_t)len)))
2766 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2773 * Reset the byte_count to the amount of bytes
2774 * we just flushed. len will be added a few lines below,
2775 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2777 byte_count = (size_t)flushed;
2779 content_filter = NULL;
2785 if (write_socket(csp->cfd, buf, (size_t)len))
2787 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2791 byte_count += (size_t)len;
2796 /* we're still looking for the end of the
2797 * server's header ... (does that make header
2798 * parsing an "out of body experience" ?
2802 * buffer up the data we just read. If that fails,
2803 * there's little we can do but send our static
2804 * out-of-memory page.
2806 if (add_to_iob(csp, buf, len))
2808 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2809 rsp = cgi_error_memory();
2810 send_crunch_response(csp, rsp);
2815 /* Convert iob into something sed() can digest */
2816 if (JB_ERR_PARSE == get_server_headers(csp))
2820 /* Well, we tried our MS IIS/5
2821 * hack and it didn't work.
2822 * The header is incomplete
2823 * and there isn't anything
2824 * we can do about it.
2826 log_error(LOG_LEVEL_INFO,
2827 "MS IIS5 hack didn't produce valid headers.");
2832 /* Since we have to wait for
2833 * more from the server before
2834 * we can parse the headers
2835 * we just continue here.
2841 /* Did we actually get anything? */
2842 if (NULL == csp->headers->first)
2844 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2845 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2846 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2847 free_http_request(http);
2851 assert(csp->headers->first->str);
2853 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2854 strncmpic(csp->headers->first->str, "ICY", 3))
2857 * It doesn't look like a HTTP (or Shoutcast) response:
2858 * tell the client and log the problem.
2860 if (strlen(csp->headers->first->str) > 30)
2862 csp->headers->first->str[30] = '\0';
2864 log_error(LOG_LEVEL_ERROR,
2865 "Invalid server or forwarder response. Starts with: %s",
2866 csp->headers->first->str);
2867 log_error(LOG_LEVEL_CLF,
2868 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2869 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2870 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2871 free_http_request(http);
2876 * We have now received the entire server header,
2877 * filter it and send the result to the client
2879 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2881 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2883 hdr = list_to_text(csp->headers);
2886 /* FIXME Should handle error properly */
2887 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2890 if (crunch_response_triggered(csp, crunchers_light))
2893 * One of the tags created by a server-header
2894 * tagger triggered a crunch. We already
2895 * delivered the crunch response to the client
2896 * and are done here after cleaning up.
2901 /* Buffer and pcrs filter this if appropriate. */
2903 if (!http->ssl) /* We talk plaintext */
2905 content_filter = get_filter_function(csp);
2908 * Only write if we're not buffering for content modification
2910 if (!content_filter)
2912 /* write the server's (modified) header to
2913 * the client (along with anything else that
2914 * may be in the buffer)
2917 if (write_socket(csp->cfd, hdr, strlen(hdr))
2918 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2920 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2922 /* the write failed, so don't bother
2923 * mentioning it to the client...
2924 * it probably can't hear us anyway.
2930 byte_count += (size_t)len;
2933 /* we're finished with the server's header */
2938 /* If this was a MS IIS/5 hack then it means
2939 * the server has already closed the
2940 * connection. Nothing more to read. Time
2945 log_error(LOG_LEVEL_INFO,
2946 "Closed server connection detected with MS IIS5 hack enabled.");
2953 return; /* huh? we should never get here */
2956 if (csp->content_length == 0)
2959 * If Privoxy didn't recalculate the
2960 * Content-Lenght, byte_count is still
2963 csp->content_length = byte_count;
2966 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2967 csp->ip_addr_str, http->ocmd, csp->content_length);
2971 /*********************************************************************
2975 * Description : This is little more than chat. We only "serve" to
2976 * to close any socket that chat may have opened.
2979 * 1 : csp = Current client state (buffers, headers, etc...)
2983 *********************************************************************/
2985 void serve(struct client_state *csp)
2986 #else /* ifndef AMIGA */
2987 static void serve(struct client_state *csp)
2988 #endif /* def AMIGA */
2991 close_socket(csp->cfd);
2993 if (csp->sfd != JB_INVALID_SOCKET)
2995 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2996 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2998 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
3002 forget_connection(csp->sfd);
3003 close_socket(csp->sfd);
3006 close_socket(csp->sfd);
3007 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3010 csp->flags &= ~CSP_FLAG_ACTIVE;
3016 /*********************************************************************
3018 * Function : server_thread
3020 * Description : We only exist to call `serve' in a threaded environment.
3023 * 1 : data = Current client state (buffers, headers, etc...)
3025 * Returns : Always 0.
3027 *********************************************************************/
3028 static int32 server_thread(void *data)
3030 serve((struct client_state *) data);
3037 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3038 /*********************************************************************
3042 * Description : Print usage info & exit.
3044 * Parameters : Pointer to argv[0] for identifying ourselves
3048 *********************************************************************/
3049 static void usage(const char *myname)
3051 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3055 #endif /* defined(unix) */
3058 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3059 #endif /* defined(unix) */
3060 "[--version] [configfile]\n"
3061 "Aborting\n", myname);
3066 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3069 #ifdef MUTEX_LOCKS_AVAILABLE
3070 /*********************************************************************
3072 * Function : privoxy_mutex_lock
3074 * Description : Locks a mutex.
3077 * 1 : mutex = The mutex to lock.
3079 * Returns : Void. May exit in case of errors.
3081 *********************************************************************/
3082 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3084 #ifdef FEATURE_PTHREAD
3085 int err = pthread_mutex_lock(mutex);
3088 if (mutex != &log_mutex)
3090 log_error(LOG_LEVEL_FATAL,
3091 "Mutex locking failed: %s.\n", strerror(err));
3096 EnterCriticalSection(mutex);
3097 #endif /* def FEATURE_PTHREAD */
3101 /*********************************************************************
3103 * Function : privoxy_mutex_unlock
3105 * Description : Unlocks a mutex.
3108 * 1 : mutex = The mutex to unlock.
3110 * Returns : Void. May exit in case of errors.
3112 *********************************************************************/
3113 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3115 #ifdef FEATURE_PTHREAD
3116 int err = pthread_mutex_unlock(mutex);
3119 if (mutex != &log_mutex)
3121 log_error(LOG_LEVEL_FATAL,
3122 "Mutex unlocking failed: %s.\n", strerror(err));
3127 LeaveCriticalSection(mutex);
3128 #endif /* def FEATURE_PTHREAD */
3132 /*********************************************************************
3134 * Function : privoxy_mutex_init
3136 * Description : Prepares a mutex.
3139 * 1 : mutex = The mutex to initialize.
3141 * Returns : Void. May exit in case of errors.
3143 *********************************************************************/
3144 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3146 #ifdef FEATURE_PTHREAD
3147 int err = pthread_mutex_init(mutex, 0);
3150 printf("Fatal error. Mutex initialization failed: %s.\n",
3155 InitializeCriticalSection(mutex);
3156 #endif /* def FEATURE_PTHREAD */
3158 #endif /* def MUTEX_LOCKS_AVAILABLE */
3160 /*********************************************************************
3162 * Function : initialize_mutexes
3164 * Description : Prepares mutexes if mutex support is available.
3168 * Returns : Void, exits in case of errors.
3170 *********************************************************************/
3171 static void initialize_mutexes(void)
3173 #ifdef MUTEX_LOCKS_AVAILABLE
3175 * Prepare global mutex semaphores
3177 privoxy_mutex_init(&log_mutex);
3178 privoxy_mutex_init(&log_init_mutex);
3179 privoxy_mutex_init(&connection_reuse_mutex);
3182 * XXX: The assumptions below are a bit naive
3183 * and can cause locks that aren't necessary.
3185 * For example older FreeBSD versions (< 6.x?)
3186 * have no gethostbyname_r, but gethostbyname is
3189 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3190 privoxy_mutex_init(&resolver_mutex);
3191 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3193 * XXX: should we use a single mutex for
3194 * localtime() and gmtime() as well?
3196 #ifndef HAVE_GMTIME_R
3197 privoxy_mutex_init(&gmtime_mutex);
3198 #endif /* ndef HAVE_GMTIME_R */
3200 #ifndef HAVE_LOCALTIME_R
3201 privoxy_mutex_init(&localtime_mutex);
3202 #endif /* ndef HAVE_GMTIME_R */
3205 privoxy_mutex_init(&rand_mutex);
3206 #endif /* ndef HAVE_RANDOM */
3207 #endif /* def MUTEX_LOCKS_AVAILABLE */
3211 /*********************************************************************
3215 * Description : Load the config file and start the listen loop.
3216 * This function is a lot more *sane* with the `load_config'
3217 * and `listen_loop' functions; although it stills does
3218 * a *little* too much for my taste.
3221 * 1 : argc = Number of parameters (including $0).
3222 * 2 : argv = Array of (char *)'s to the parameters.
3224 * Returns : 1 if : can't open config file, unrecognized directive,
3225 * stats requested in multi-thread mode, can't open the
3226 * log file, can't open the jar file, listen port is invalid,
3227 * any load fails, and can't bind port.
3229 * Else main never returns, the process must be signaled
3230 * to terminate execution. Or, on Windows, use the
3231 * "File", "Exit" menu option.
3233 *********************************************************************/
3235 int real_main(int argc, const char *argv[])
3237 int main(int argc, const char *argv[])
3241 unsigned int random_seed;
3243 struct passwd *pw = NULL;
3244 struct group *grp = NULL;
3247 char *pre_chroot_nslookup_to_load_resolver = NULL;
3254 #if !defined(_WIN32)
3262 * Parse the command line arguments
3264 * XXX: simply printing usage information in case of
3265 * invalid arguments isn't particularly user friendly.
3267 while (++argc_pos < argc)
3270 /* Check to see if the service must be installed or uninstalled */
3271 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3273 const char *pName = argv[argc_pos] + 9;
3276 exit( (install_service(pName)) ? 0 : 1 );
3278 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3280 const char *pName = argv[argc_pos] + 11;
3283 exit((uninstall_service(pName)) ? 0 : 1);
3285 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3287 bRunAsService = TRUE;
3288 w32_set_service_cwd();
3289 atexit(w32_service_exit_notify);
3292 #endif /* defined(_WIN32) */
3295 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3297 if (strcmp(argv[argc_pos], "--help") == 0)
3302 else if(strcmp(argv[argc_pos], "--version") == 0)
3304 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3310 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3315 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3317 if (++argc_pos == argc) usage(argv[0]);
3318 pidfile = strdup(argv[argc_pos]);
3321 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3323 if (++argc_pos == argc) usage(argv[argc_pos]);
3325 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3328 if (NULL == (grp = getgrnam(p)))
3330 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3334 if (NULL == (pw = getpwnam(argv[argc_pos])))
3336 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3339 if (p != NULL) *--p = '\0';
3342 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3344 if (++argc_pos == argc) usage(argv[0]);
3345 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3348 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3352 #endif /* defined(unix) */
3354 else if (argc_pos + 1 != argc)
3357 * This is neither the last command line
3358 * option, nor was it recognized before,
3359 * therefore it must be invalid.
3365 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3367 configfile = argv[argc_pos];
3370 } /* -END- while (more arguments) */
3373 if ( *configfile != '/' )
3375 char cwd[BUFFER_SIZE];
3377 size_t abs_file_size;
3379 /* make config-filename absolute here */
3380 if (NULL == getcwd(cwd, sizeof(cwd)))
3382 perror("failed to get current working directory");
3387 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3388 basedir = strdup(cwd);
3390 if (NULL == basedir ||
3391 NULL == (abs_file = malloc(abs_file_size)))
3393 perror("malloc failed");
3396 strlcpy(abs_file, basedir, abs_file_size);
3397 strlcat(abs_file, "/", abs_file_size );
3398 strlcat(abs_file, configfile, abs_file_size);
3399 configfile = abs_file;
3401 #endif /* defined unix */
3405 clients->next = NULL;
3407 /* XXX: factor out initialising after the next stable release. */
3410 #elif defined(_WIN32)
3414 /* Prepare mutexes if supported and necessary. */
3415 initialize_mutexes();
3417 /* Enable logging until further notice. */
3418 init_log_module(Argv[0]);
3420 random_seed = (unsigned int)time(NULL);
3422 srandom(random_seed);
3425 #endif /* ifdef HAVE_RANDOM */
3428 * Unix signal handling
3430 * Catch the abort, interrupt and terminate signals for a graceful exit
3431 * Catch the hangup signal so the errlog can be reopened.
3432 * Ignore the broken pipe signals (FIXME: Why?)
3434 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3437 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3438 const int ignored_signals[] = { SIGPIPE, 0 };
3440 for (idx = 0; catched_signals[idx] != 0; idx++)
3442 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3443 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3445 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3446 #endif /* ifdef sun */
3448 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3452 for (idx = 0; ignored_signals[idx] != 0; idx++)
3454 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3456 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3461 #else /* ifdef _WIN32 */
3462 # ifdef _WIN_CONSOLE
3464 * We *are* in a windows console app.
3465 * Print a verbose messages about FAQ's and such
3467 printf("%s", win32_blurb);
3468 # endif /* def _WIN_CONSOLE */
3469 #endif /* def _WIN32 */
3472 /* Initialize the CGI subsystem */
3473 cgi_init_error_messages();
3476 * If runnig on unix and without the --nodaemon
3477 * option, become a daemon. I.e. fork, detach
3478 * from tty and get process group leadership
3491 if ( pid < 0 ) /* error */
3496 else if ( pid != 0 ) /* parent */
3501 * must check for errors
3502 * child died due to missing files aso
3505 wpid = waitpid( pid, &status, WNOHANG );
3514 /* Should be more portable, but not as well tested */
3519 #else /* ndef __FreeBSD__ */
3521 #endif /* ndef __FreeBSD__ */
3522 fd = open("/dev/tty", O_RDONLY);
3525 /* no error check here */
3526 ioctl( fd, TIOCNOTTY,0 );
3531 * stderr (fd 2) will be closed later on, when the
3532 * log file has been parsed.
3539 } /* -END- if (!no_daemon) */
3542 * As soon as we have written the PID file, we can switch
3543 * to the user and group ID indicated by the --user option
3549 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3551 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3557 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3559 /* Read the time zone file from /etc before doing chroot. */
3561 if (NULL != pre_chroot_nslookup_to_load_resolver
3562 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3564 /* Initialize resolver library. */
3565 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3567 if (chroot(pw->pw_dir) < 0)
3569 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3573 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3576 if (setuid(pw->pw_uid))
3578 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3582 char putenv_dummy[64];
3584 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3585 if (putenv(putenv_dummy) != 0)
3587 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3590 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3591 if (putenv(putenv_dummy) != 0)
3593 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3599 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3602 #endif /* defined unix */
3605 /* This will be FALSE unless the command line specified --service
3609 /* Yup, so now we must attempt to establish a connection
3610 * with the service dispatcher. This will only work if this
3611 * process was launched by the service control manager to
3612 * actually run as a service. If this isn't the case, i've
3613 * known it take around 30 seconds or so for the call to return.
3616 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3617 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3619 /* Service has run, and at this point is now being stopped, so just return */
3624 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3626 /* An error occurred. Usually it's because --service was wrongly specified
3627 * and we were unable to connect to the Service Control Dispatcher because
3628 * it wasn't expecting us and is therefore not listening.
3630 * For now, just continue below to call the listen_loop function.
3633 #endif /* def _WIN32 */
3643 /*********************************************************************
3645 * Function : bind_port_helper
3647 * Description : Bind the listen port. Handles logging, and aborts
3651 * 1 : config = Privoxy configuration. Specifies port
3654 * Returns : Port that was opened.
3656 *********************************************************************/
3657 static jb_socket bind_port_helper(struct configuration_spec * config)
3662 if (config->haddr == NULL)
3664 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3669 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3670 config->hport, config->haddr);
3673 result = bind_port(config->haddr, config->hport, &bfd);
3680 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3681 "There may be another Privoxy or some other "
3682 "proxy running on port %d",
3683 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3684 config->hport, config->hport);
3687 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3688 "The hostname is not resolvable",
3689 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3692 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3693 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3696 /* shouldn't get here */
3697 return JB_INVALID_SOCKET;
3700 config->need_bind = 0;
3707 /* Without this simple workaround we get this compiler warning from _beginthread
3708 * warning C4028: formal parameter 1 different from declaration
3710 void w32_service_listen_loop(void *p)
3714 #endif /* def _WIN32 */
3717 /*********************************************************************
3719 * Function : listen_loop
3721 * Description : bind the listen port and enter a "FOREVER" listening loop.
3727 *********************************************************************/
3728 static void listen_loop(void)
3730 struct client_state *csp = NULL;
3732 struct configuration_spec * config;
3734 config = load_config();
3736 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3738 * XXX: Should be relocated once it no
3739 * longer needs to emit log messages.
3741 initialize_reusable_connections();
3742 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3744 bfd = bind_port_helper(config);
3746 #ifdef FEATURE_GRACEFUL_TERMINATION
3747 while (!g_terminate)
3752 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3753 while (waitpid(-1, NULL, WNOHANG) > 0)
3755 /* zombie children */
3757 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3760 * Free data that was used by died threads
3766 * Re-open the errlog after HUP signal
3768 if (received_hup_signal)
3770 init_error_log(Argv[0], config->logfile);
3771 received_hup_signal = 0;
3775 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3777 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3781 csp->flags |= CSP_FLAG_ACTIVE;
3782 csp->sfd = JB_INVALID_SOCKET;
3784 csp->config = config = load_config();
3786 if ( config->need_bind )
3789 * Since we were listening to the "old port", we will not see
3790 * a "listen" param change until the next IJB request. So, at
3791 * least 1 more request must be made for us to find the new
3792 * setting. I am simply closing the old socket and binding the
3795 * Which-ever is correct, we will serve 1 more page via the
3796 * old settings. This should probably be a "show-proxy-args"
3797 * request. This should not be a so common of an operation
3798 * that this will hurt people's feelings.
3803 bfd = bind_port_helper(config);
3806 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3808 if (!accept_connection(csp, bfd))
3810 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3823 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3826 #ifdef FEATURE_TOGGLE
3827 if (global_toggle_state)
3828 #endif /* def FEATURE_TOGGLE */
3830 csp->flags |= CSP_FLAG_TOGGLED_ON;
3833 if (run_loader(csp))
3835 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3836 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3840 if (block_acl(NULL,csp))
3842 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3843 close_socket(csp->cfd);
3847 #endif /* def FEATURE_ACL */
3849 /* add it to the list of clients */
3850 csp->next = clients->next;
3851 clients->next = csp;
3853 if (config->multi_threaded)
3857 /* this is a switch () statment in the C preprocessor - ugh */
3858 #undef SELECTED_ONE_OPTION
3860 /* Use Pthreads in preference to native code */
3861 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3862 #define SELECTED_ONE_OPTION
3864 pthread_t the_thread;
3865 pthread_attr_t attrs;
3867 pthread_attr_init(&attrs);
3868 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3869 errno = pthread_create(&the_thread, &attrs,
3870 (void * (*)(void *))serve, csp);
3871 child_id = errno ? -1 : 0;
3872 pthread_attr_destroy(&attrs);
3876 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3877 #define SELECTED_ONE_OPTION
3878 child_id = _beginthread(
3879 (void (*)(void *))serve,
3884 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3885 #define SELECTED_ONE_OPTION
3886 child_id = _beginthread(
3887 (void(* _Optlink)(void*))serve,
3893 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3894 #define SELECTED_ONE_OPTION
3896 thread_id tid = spawn_thread
3897 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3899 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3901 child_id = (int) tid;
3910 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3911 #define SELECTED_ONE_OPTION
3912 csp->cfd = ReleaseSocket(csp->cfd, -1);
3915 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3916 NP_Output, Output(),
3917 NP_CloseOutput, FALSE,
3918 NP_Name, (ULONG)"privoxy child",
3922 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3923 NP_Output, Output(),
3924 NP_CloseOutput, FALSE,
3925 NP_Name, (ULONG)"privoxy child",
3926 NP_StackSize, 200*1024,
3932 ((struct Task *)child_id)->tc_UserData = csp;
3933 Signal((struct Task *)child_id, SIGF_SINGLE);
3938 #if !defined(SELECTED_ONE_OPTION)
3941 /* This block is only needed when using fork().
3942 * When using threads, the server thread was
3943 * created and run by the call to _beginthread().
3945 if (child_id == 0) /* child */
3948 #ifdef FEATURE_TOGGLE
3949 int inherited_toggle_state = global_toggle_state;
3950 #endif /* def FEATURE_TOGGLE */
3955 * If we've been toggled or we've blocked the request, tell Mom
3958 #ifdef FEATURE_TOGGLE
3959 if (inherited_toggle_state != global_toggle_state)
3961 rc |= RC_FLAG_TOGGLED;
3963 #endif /* def FEATURE_TOGGLE */
3965 #ifdef FEATURE_STATISTICS
3966 if (csp->flags & CSP_FLAG_REJECTED)
3968 rc |= RC_FLAG_BLOCKED;
3970 #endif /* ndef FEATURE_STATISTICS */
3974 else if (child_id > 0) /* parent */
3976 /* in a fork()'d environment, the parent's
3977 * copy of the client socket and the CSP
3981 #if !defined(_WIN32) && !defined(__CYGWIN__)
3983 wait( &child_status );
3986 * Evaluate child's return code: If the child has
3987 * - been toggled, toggle ourselves
3988 * - blocked its request, bump up the stats counter
3991 #ifdef FEATURE_TOGGLE
3992 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3994 global_toggle_state = !global_toggle_state;
3996 #endif /* def FEATURE_TOGGLE */
3998 #ifdef FEATURE_STATISTICS
4000 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4004 #endif /* def FEATURE_STATISTICS */
4006 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4007 close_socket(csp->cfd);
4008 csp->flags &= ~CSP_FLAG_ACTIVE;
4012 #undef SELECTED_ONE_OPTION
4013 /* end of cpp switch () */
4015 if (child_id < 0) /* failed */
4017 char buf[BUFFER_SIZE];
4019 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
4021 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
4023 write_socket(csp->cfd, buf, strlen(buf));
4024 close_socket(csp->cfd);
4025 csp->flags &= ~CSP_FLAG_ACTIVE;
4036 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4038 /* Clean up. Aim: free all memory (no leaks) */
4039 #ifdef FEATURE_GRACEFUL_TERMINATION
4041 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4043 unload_current_config_file();
4044 unload_current_actions_file();
4045 unload_current_re_filterfile();
4046 #ifdef FEATURE_TRUST
4047 unload_current_trust_file();
4050 if (config->multi_threaded)
4057 } while ((clients->next != NULL) && (--i > 0));
4061 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4072 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4073 /* Cleanup - remove taskbar icon etc. */
4078 #endif /* FEATURE_GRACEFUL_TERMINATION */