1 const char jcc_rcs[] = "$Id: jcc.c,v 1.194 2008/10/12 18:35:18 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 1.194 2008/10/12 18:35:18 fabiankeil
37 * The last commit was a bit too ambitious, apparently the content
38 * length adjustment is only necessary if we aren't buffering.
40 * Revision 1.193 2008/10/12 15:57:35 fabiankeil
41 * Fix content length calculation if we read headers
42 * and the start of the body at once. Now that we have
43 * FEATURE_CONNECTION_KEEP_ALIVE, it actually matters.
45 * Revision 1.192 2008/10/11 18:19:14 fabiankeil
46 * Even more chat() cosmetics.
48 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
49 * Reformat some comments in chat().
51 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
52 * In case of chunk-encoded content, stop reading if
53 * the buffer looks like it ends with the last chunk.
55 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
56 * Let server_response_is_complete() deal properly with
57 * content that is neither buffered nor read all at once.
59 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
60 * Flush work-in-progress changes to keep outgoing connections
61 * alive where possible. Incomplete and mostly #ifdef'd out.
63 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
64 * Add mutex lock support for _WIN32.
66 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
67 * Prepare for critical sections on Windows by adding a
68 * layer of indirection before the pthread mutex functions.
70 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
71 * Remove FEATURE_COOKIE_JAR.
73 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
74 * - Silence LLVM/Clang complaint.
75 * - Make received_hup_signal static.
76 * - Hide definitions for basedir, pidfile and received_hup_signal
77 * from __EMX__ as they only seem to be used in case of #ifdef unix.
79 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
80 * Accept Shoutcast responses again. Problem reported
81 * and fix suggested by Stefan in #2062860.
83 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
84 * Fix possible NULL-pointer dereference reported
85 * by din_a4 in #2003937. Pointy hat to me.
87 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
88 * Streamline sed()'s prototype and declare
89 * the header parse and add structures static.
91 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
92 * - Mark csp as immutable for send_crunch_response().
93 * - Fix comment spelling.
95 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
96 * Factor update_server_headers() out of sed(), ditch the
97 * first_run hack and make server_patterns_light static.
99 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
100 * Don't provide get_header() with the whole client state
101 * structure when it only needs access to csp->iob.
103 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
104 * Make the "read the rest of the headers" loop a bit more readable.
106 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
107 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
108 * - Remove useless comment.
110 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
111 * Fix comment grammar.
113 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
114 * Remove the pointless buffer in client_protocol_is_unsupported().
116 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
117 * Least-effort fix for bug #1821930 (reported by Lee):
118 * If the response doesn't look like HTTP,
119 * tell the client and log the problem.
121 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
122 * Don't pass the whole csp structure to flush_socket()
123 * when it only needs a file descriptor and a buffer.
125 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
126 * Remove kill-popups action.
128 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
129 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
131 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
132 * Remove the treat-forbidden-connects-like-blocks action. We now
133 * use the "blocked" page for forbidden CONNECT requests by default.
135 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
136 * Also use shiny new connect_port_is_forbidden() in jcc.c.
138 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
139 * Rename url_actions() to get_url_actions() and let it
140 * use the standard parameter ordering.
142 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
143 * Let forward_url() use the standard parameter ordering
144 * and mark its second parameter immutable.
146 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
147 * Remove the "Listening ... for local connections only" log message.
148 * Whether or not remote connections are able to reach Privoxy is up
149 * to the operating system.
151 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
152 * Prevent the log messages for CONNECT requests to unacceptable
153 * ports from printing the limit-connect argument as [null] if
154 * limit-connect hasn't been explicitly enabled.
156 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
157 * Make sure all console-mode apps get a usage() instance
159 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
160 * Reword NO_SERVER_DATA_RESPONSE to make it harder
161 * to misunderstand what the message is all about.
163 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
164 * Unbreak trustfile which previously didn't work without
165 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
167 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
168 * Plug memory leak. Spotted by Valgrind, triggered by
169 * Privoxy-Regression-Test feeding proxyfuzz.py.
171 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
172 * In the HTTP snipplets, refer to the client as client.
174 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
175 * Emit a log message when activating the MS IIS5 hack.
177 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
178 * Log the "weak randomization factor" warning only
179 * once for mingw32 and provide some more details.
181 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
182 * Initialize log module after initializing mutexes, future
183 * deadlocks in that code should now work cross-platform.
185 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
186 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
187 * Reported by Bert van Leeuwen in BR#1818808.
189 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
190 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
192 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
193 * When in daemon mode, close stderr after the configuration file has been
194 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
196 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
197 * - Fix a crash when parsing invalid requests whose first header
198 * is rejected by get_header(). Regression (re?)introduced
199 * in r1.143 by yours truly.
200 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
201 * client_cookie_adder() to make sure send-vanilla-wafer can be
202 * controlled through tags (and thus regression-tested).
204 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
205 * - Move get_filter_function() from jcc.c to filters.c
206 * so the filter functions can be static.
207 * - Don't bother filtering body-less responses.
209 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
210 * Execute content filters through execute_content_filter().
212 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
213 * Initialize req to NULL to make sure it's defined if the
214 * first read_socket() call fails. Reported by icmp30.
216 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
217 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
218 * extensive comments moved to user manual.
220 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
221 * Don't crash if a broken header filter wiped out the request line.
223 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
224 * Fix byte_count calculation in case of flushes
225 * and don't parse the server headers a second time.
227 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
228 * - If there's a connection problem after we already forwarded
229 * parts of the original content, just hang up. Fixes BR#1776724.
230 * - Fix warnings about unused code on mingw32.
231 * - In case of flushes, calculate the byte count
232 * less incorrectly (I think).
234 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
235 * Add some more prototypes for static functions.
237 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
238 * Comment out request_contains_null_bytes() until it's used again.
240 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
241 * #1763173 from Stefan Huehner: s@const static@static const@
242 * and declare some more functions static.
244 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
245 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
246 * and explain why they were rejected in the first place.
247 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
248 * CONNECT requests. The request line was missing.
249 * - Add two more XXX reminders as we don't have enough already.
251 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
252 * As Hal noticed, checking dispatch_cgi() as the last cruncher
253 * looks like a bug if CGI requests are blocked unintentionally,
254 * so don't do it unless the user enabled the new config option
255 * "allow-cgi-request-crunching".
257 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
258 * - Allow to rewrite the request destination behind the client's back.
259 * - Turn the weird-looking unconditional for loop that
260 * reads the client request into a conditional while loop.
261 * Move the stuff that only runs once out of the loop.
262 * - Move parts of chat(), server_content_type() and the
263 * necessary stuff to fix BR#1750917 into get_filter_function().
265 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
266 * Temporary workaround for BR#1730105.
268 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
269 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
270 * deadlocks and crashes on OpenBSD and possibly other OS with neither
271 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
272 * Thanks to Ralf Horstmann for report and solution.
274 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
275 * Add forward-override{} to change the forwarding settings through
276 * action sections. This is mainly interesting to forward different
277 * clients differently (for example based on User-Agent or request
280 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
281 * - Let usage() mention the --chroot parameter.
282 * - Use read_socket() consistently and always leave
283 * the last buffer byte alone, even in cases where
284 * null termination (currently) doesn't matter.
286 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
287 * - Fix config file loading on Unix if no config file is specified.
288 * Since r1.97 Privoxy would always interpret the last argument as
289 * config file, even if it's a valid command line option.
290 * - Abort in case of unrecognized command line options. Closes #1719696.
291 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
292 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
294 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
295 * - Don't rerun crunchers that only depend on the request URL.
296 * - Don't count redirects and CGI requests as "blocked requests".
298 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
299 * Support crunching based on tags created by server-header taggers.
301 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
302 * Make HTTP snippets static (again). Add a Content-Type for those
303 * with content so the browser doesn't guess it based on the URL.
305 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
306 * Move crunching and request line rebuilding out of chat().
308 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
309 * Introduce tags as alternative way to specify which
310 * actions apply to a request. At the moment tags can be
311 * created based on client and server headers.
313 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
314 * Don't CLF-log CONNECT requests twice.
316 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
317 * Log the source address for ACL-related connection drops.
319 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
320 * New config option: enforce-blocks.
322 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
323 * - Move null byte check into separate function.
324 * - Don't confuse the client with error pages
325 * if a CONNECT request was already confirmed.
327 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
328 * Speed up NULL byte escaping and only log the complete
329 * NULL byte requests with header debugging enabled.
331 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
332 * Answer requests that contain NULL bytes with
333 * a custom response instead of waiting for more
334 * data until the client eventually hangs up.
336 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
337 * - Move delivery and logging of crunched responses
338 * from chat() into send_crunch_response().
339 * - Display the reason for generating http_responses.
340 * - Log the content length for LOG_LEVEL_CLF correctly
341 * (still incorrect for some fixed responses).
342 * - Reword an incorrect comment about
343 * treat-forbidden-connects-like-blocks violating
345 * - Add some log messages.
347 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
348 * Move mutex initialization into separate
349 * function and exit in case of errors.
351 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
352 * - Start to reduce chat()'s line count and move
353 * parts of it into separate functions.
354 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
356 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
357 * - Add Proxy-Agent header to HTTP snippets that are
358 * supposed to reach HTTP clients only.
359 * - Made a few CONNECT log messages more descriptive.
360 * - Catch completely empty server responses (as seen
361 * with Tor's fake ".noconnect" top level domain).
362 * - Use shiny new "forwarding-failed" template for socks errors.
364 * Revision 1.118 2007/01/07 07:43:43 joergs
365 * AmigaOS4 support added.
367 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
368 * Added config option accept-intercepted-requests
369 * and disabled it by default.
371 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
372 * Reverted parts of my last commit
373 * to keep error handling working.
375 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
376 * Fixed gcc43 conversion warnings.
378 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
379 * Fix -pedantic ISO C warning about converting
380 * from function pointer to object pointer.
382 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
383 * Silence compiler warning I introduced with my last commit.
385 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
386 * Mutex protect rand() if POSIX threading
387 * is used, warn the user if that's not possible
388 * and stop using it on _WIN32 where it could
391 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
392 * Don't prevent core dumps by catching SIGABRT.
393 * It's rude and makes debugging unreasonable painful.
395 * Revision 1.110 2006/12/13 14:52:53 etresoft
396 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
398 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
399 * Privoxy is now able to run as intercepting
400 * proxy in combination with any packet filter
401 * that does the port redirection. The destination
402 * is extracted from the "Host:" header which
403 * should be available for nearly all requests.
405 * Moved HTTP snipplets into jcc.c.
406 * Added error message for gopher proxy requests.
408 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
409 * Only unlink the pidfile if it's actually used.
411 * Change order of interception checks to make
412 * it possible to block or redirect requests for
415 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
416 * Make pthread mutex locking more generic. Instead of
417 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
418 * and use mutex locking unless there is an _r function
419 * available. Better safe than sorry.
421 * Fixes "./configure --disable-pthread" and should result
422 * in less threading-related problems on pthread-using platforms,
423 * but it still doesn't fix BR#1122404.
425 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
426 * Move pthread.h inclusion from jcc.c to jcc.h.
427 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
429 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
430 * Don't exit after receiving the second SIGHUP on Solaris.
432 * Fixes BR 1052235, but the same problem may exist on other
433 * systems. Once 3.0.6 is out we should use sigset()
434 * where available and see if it breaks anything.
436 * Revision 1.104 2006/09/23 13:26:38 roro
437 * Replace TABs by spaces in source code.
439 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
440 * Fix +redirect{}. Didn't work with -fast-redirects.
442 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
443 * Respond with 400 and a short text message
444 * if the client tries to use Privoxy as FTP proxy.
446 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
447 * Make number of retries in case of forwarded-connect problems
448 * a config file option (forwarded-connect-retries) and use 0 as
451 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
452 * Set random(3) seed.
454 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
455 * Follow the OpenBSD port's lead and protect the resolve
456 * functions on OpenBSD as well.
458 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
459 * --user fix. Only use the user as group if no group is specified.
460 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
462 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
463 * Windows service (re-)integration
467 * --install[:service_name]
468 * --uninstall[:service_name]
471 * They work as follows:
472 * --install will create a service for you and then terminate.
473 * By default the service name will be "privoxy" (without the quotes).
474 * However you can run multiple services if you wish, just by adding
475 * a colon and then a name (no spaces).
477 * --uninstall follows the exact same rules a --install.
479 * --service is used when the program is executed by the service
480 * control manager, and in normal circumstances would never be
481 * used as a command line argument.
483 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
484 * Windows service integration
486 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
487 * Incorporate Fabian Keil's patch work:
488 http://www.fabiankeil.de/sourcecode/privoxy/
490 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
491 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
492 * with what was really the latest development (the v_3_0_branch branch)
494 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
495 * Thanks to Jindrich Makovicka for a race condition fix for the log
496 * file. The race condition remains for non-pthread implementations.
497 * Reference patch #1175720.
499 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
500 * Add the ability to check jpeg images for invalid
501 * lengths of comment blocks. Defensive strategy
502 * against the exploit:
503 * Microsoft Security Bulletin MS04-028
504 * Buffer Overrun in JPEG Processing (GDI+) Could
505 * Allow Code Execution (833987)
506 * Enabled with +inspect-jpegs in actions files.
508 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
509 * - Fixed usage info for non-unix platforms
510 * - Fixed small cmdline parsing bug
512 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
513 * Diagnostics: Now preserve the returncode of pthread_create
514 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
516 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
517 * No longer ignore SIGCHLD. Fixes bug #769381
519 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
520 * Close jarfile on graceful exit, remove stray line
522 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
523 * Cosmetics: Killed a warning, a typo and an allocation left at exit
525 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
526 * No longer rely on non-POSIX.1 extensions of getcwd().
529 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
530 * Replaced setenv() by posix-compliant putenv()
531 * Thanks to Neil McCalden (nmcc AT users.sf.net).
533 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
534 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
536 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
537 * Clean-up and extension of improvements for forked mode:
538 * - Child's return code now consists of flags RC_FLAG_*
539 * - Reporting toggle to parent now properly #ifdef'ed
540 * - Children now report blocking to parent. This enables
541 * statistics in forked mode
543 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
544 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
545 * Children now report having been toggled through _exit(17), parents
546 * watch for that code and toggle themselves if found.
548 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
549 * Wrapping all *_r functions (the non-_r versions of them) with
550 * mutex semaphores for OSX. Hopefully this will take care of all
551 * of those pesky crash reports.
553 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
554 * Fixed two mostly harmless mem leaks
556 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
557 * Fix: Head of global clients list now initialized to NULL
559 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
560 * Added basic support for OPTIONS and TRACE HTTP methods:
561 * - New interceptor direct_response() added in chat().
562 * - sed() moved to earlier in the process, so that the
563 * Host: header is evaluated before actions and forwarding
566 * Revision 1.92 2002/05/08 16:00:46 oes
567 * Chat's buffer handling:
568 * - Fixed bug with unchecked out-of-mem conditions
569 * while reading client request & server headers
570 * - No longer predict if the buffer limit will be exceeded
571 * in the next read -- check add_to_iob's new
572 * return code. If buffer couldn't be extended
573 * (policy or out-of-mem) while
574 * - reading from client: abort
575 * - reading server headers: send error page
576 * - buffering server body for filter: flush,
577 * and if that fails: send error page
579 * Revision 1.91 2002/04/08 20:35:58 swa
582 * Revision 1.90 2002/04/02 14:57:28 oes
583 * Made sending wafers independent of FEATURE_COOKIE_JAR
585 * Revision 1.89 2002/03/31 17:18:59 jongfoster
586 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
588 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
589 * More compiler warning message maintenance
591 * Revision 1.87 2002/03/26 22:29:54 swa
592 * we have a new homepage!
594 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
595 * Workaround for closing the jarfile before load_config() comes around again
597 * Revision 1.85 2002/03/24 15:23:33 jongfoster
600 * Revision 1.84 2002/03/24 13:25:43 swa
601 * name change related issues
603 * Revision 1.83 2002/03/16 23:54:06 jongfoster
604 * Adding graceful termination feature, to help look for memory leaks.
605 * If you enable this (which, by design, has to be done by hand
606 * editing config.h) and then go to http://i.j.b/die, then the program
607 * will exit cleanly after the *next* request. It should free all the
608 * memory that was used.
610 * Revision 1.82 2002/03/13 00:27:05 jongfoster
613 * Revision 1.81 2002/03/12 01:42:50 oes
614 * Introduced modular filters
616 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
617 * OS/2 port maintenance:
618 * - Fixed EMX build - it had decayed a little
619 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
620 * substituted a memset for now.
622 * Revision 1.79 2002/03/09 20:03:52 jongfoster
623 * - Making various functions return int rather than size_t.
624 * (Undoing a recent change). Since size_t is unsigned on
625 * Windows, functions like read_socket that return -1 on
626 * error cannot return a size_t.
628 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
629 * crashes, and also frequently caused JB to jump to 100%
630 * CPU and stay there. (Because it thought it had just
631 * read ((unsigned)-1) == 4Gb of data...)
633 * - The signature of write_socket has changed, it now simply
634 * returns success=0/failure=nonzero.
636 * - Trying to get rid of a few warnings --with-debug on
637 * Windows, I've introduced a new type "jb_socket". This is
638 * used for the socket file descriptors. On Windows, this
639 * is SOCKET (a typedef for unsigned). Everywhere else, it's
640 * an int. The error value can't be -1 any more, so it's
641 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
642 * Windows it maps to the #define INVALID_SOCKET.)
644 * - The signature of bind_port has changed.
646 * Revision 1.78 2002/03/08 21:35:04 oes
647 * Added optional group supplement to --user option. Will now use default group of user if no group given
649 * Revision 1.77 2002/03/07 03:52:06 oes
650 * - Fixed compiler warnings etc
651 * - Improved handling of failed DNS lookups
653 * Revision 1.76 2002/03/06 22:54:35 jongfoster
654 * Automated function-comment nitpicking.
656 * Revision 1.75 2002/03/06 10:02:19 oes
657 * Fixed stupid bug when --user was not given
659 * Revision 1.74 2002/03/06 00:49:31 jongfoster
660 * Fixing warning on Windows
661 * Making #ifdefs that refer to the same variable consistently
662 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
664 * Revision 1.73 2002/03/05 23:57:30 hal9
665 * Stray character 's' on line 1618 was breaking build.
667 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
668 * - Re-enable OS/2 building after new parms were added
669 * - Fix false out of memory report when resolving CGI templates when no IP
670 * address is available of failed attempt (a la no such domain)
672 * Revision 1.71 2002/03/05 18:13:56 oes
673 * Added --user option
675 * Revision 1.70 2002/03/05 04:52:42 oes
676 * Deleted non-errlog debugging code
678 * Revision 1.69 2002/03/04 23:50:00 jongfoster
679 * Splitting off bind_port() call into bind_port_helper(), with
682 * Revision 1.68 2002/03/04 20:17:32 oes
685 * Revision 1.67 2002/03/04 18:18:57 oes
686 * - Removed _DEBUG mode
687 * - Cleand up cmdline parsing
688 * - Introduced --no-daemon, --pidfile options
689 * - Cleaned up signal handling:
690 * - Terminate cleanly on INT, TERM and ABRT
691 * - Schedule logfile for re-opening on HUP
692 * - Ignore CHLD and PIPE
693 * - Leave the rest with their default handlers
694 * - Uniform handler registration
695 * - Added usage() function
696 * - Played styleguide police
698 * Revision 1.66 2002/03/03 15:06:55 oes
699 * Re-enabled automatic config reloading
701 * Revision 1.65 2002/03/03 14:49:11 oes
702 * Fixed CLF logging: Now uses client's original HTTP request
704 * Revision 1.64 2002/03/03 09:18:03 joergs
705 * Made jumbjuster work on AmigaOS again.
707 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
708 * Clean up a little CRLF unpleasantness that suddenly appeared
710 * Revision 1.62 2002/02/20 23:17:23 jongfoster
711 * Detecting some out-of memory conditions and exiting with a log message.
713 * Revision 1.61 2002/01/17 21:01:52 jongfoster
714 * Moving all our URL and URL pattern parsing code to urlmatch.c.
716 * Revision 1.60 2001/12/30 14:07:32 steudten
717 * - Add signal handling (unix)
718 * - Add SIGHUP handler (unix)
719 * - Add creation of pidfile (unix)
720 * - Add action 'top' in rc file (RH)
721 * - Add entry 'SIGNALS' to manpage
722 * - Add exit message to logfile (unix)
724 * Revision 1.59 2001/12/13 14:07:18 oes
725 * Fixed Bug: 503 error page now sent OK
727 * Revision 1.58 2001/11/30 23:37:24 jongfoster
728 * Renaming the Win32 config file to config.txt - this is almost the
729 * same as the corresponding UNIX name "config"
731 * Revision 1.57 2001/11/16 00:47:43 jongfoster
732 * Changing the tty-disconnection code to use setsid().
734 * Revision 1.56 2001/11/13 20:20:54 jongfoster
735 * Tabs->spaces, fixing a bug with missing {} around an if()
737 * Revision 1.55 2001/11/13 20:14:53 jongfoster
738 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
740 * Revision 1.54 2001/11/07 00:03:14 steudten
741 * Give reliable return value if an error
742 * occurs not just 0 with new daemon mode.
744 * Revision 1.53 2001/11/05 21:41:43 steudten
745 * Add changes to be a real daemon just for unix os.
746 * (change cwd to /, detach from controlling tty, set
747 * process group and session leader to the own process.
749 * Add some fatal-error log message for failed malloc().
750 * Add '-d' if compiled with 'configure --with-debug' to
751 * enable debug output.
753 * Revision 1.52 2001/10/26 20:11:20 jongfoster
754 * Fixing type mismatch
756 * Revision 1.51 2001/10/26 17:38:28 oes
759 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
760 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
761 * threads to call select() simultaneously. So, it's time to do a real, live,
762 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
763 * (native). Both versions will work, but using __OS2__ offers multi-threading.
765 * Revision 1.49 2001/10/23 21:41:35 jongfoster
766 * Added call to initialize the (statically-allocated of course)
767 * "out of memory" CGI response.
769 * Revision 1.48 2001/10/10 19:56:46 jongfoster
770 * Moving some code that wasn't cookie-related out of an #ifdef
773 * Revision 1.47 2001/10/10 16:44:36 oes
774 * Added CONNECT destination port limitation check
776 * Revision 1.46 2001/10/08 15:17:41 oes
777 * Re-enabled SSL forwarding
779 * Revision 1.45 2001/10/07 15:42:11 oes
780 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
782 * Moved downgrading of the HTTP version from parse_http_request to
783 * chat(), since we can't decide if it is necessary before we have
784 * determined the actions for the URL. The HTTP command is now
785 * *always* re-built so the repairs need no longer be special-cased.
787 * filter_popups now gets a csp pointer so it can raise the new
788 * CSP_FLAG_MODIFIED flag.
792 * Added configurable size limit for the IOB. If the IOB grows so
793 * large that the next read would exceed the limit, the header
794 * is generated, and the header & unfiltered buffer are flushed
795 * to the client. Chat then continues in non-buffering,
796 * non-filtering body mode.
798 * Revision 1.44 2001/10/02 18:13:57 oes
801 * Revision 1.43 2001/10/02 15:32:13 oes
802 * Moved generation of hdr
804 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
805 * Cleaning up 2 compiler warnings on OS/2.
807 * Revision 1.41 2001/09/16 17:05:14 jongfoster
808 * Removing unused #include showarg.h
810 * Revision 1.40 2001/09/16 15:41:45 jongfoster
811 * Fixing signed/unsigned comparison warning.
813 * Revision 1.39 2001/09/16 13:21:27 jongfoster
814 * Changes to use new list functions.
816 * Revision 1.38 2001/09/16 13:01:46 jongfoster
817 * Removing redundant function call that zeroed zalloc()'d memory.
819 * Revision 1.37 2001/09/10 11:12:24 oes
820 * Deleted unused variable
822 * Revision 1.36 2001/09/10 10:56:15 oes
823 * Silenced compiler warnings
825 * Revision 1.35 2001/07/31 14:44:22 oes
826 * Deleted unused size parameter from filter_popups()
828 * Revision 1.34 2001/07/30 22:08:36 jongfoster
829 * Tidying up #defines:
830 * - All feature #defines are now of the form FEATURE_xxx
831 * - Permanently turned off WIN_GUI_EDIT
832 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
834 * Revision 1.33 2001/07/29 19:32:00 jongfoster
835 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
837 * Revision 1.32 2001/07/29 18:47:05 jongfoster
838 * Adding missing #include "loadcfg.h"
840 * Revision 1.31 2001/07/29 12:17:48 oes
841 * Applied pthread fix by Paul Lieverse
843 * Revision 1.30 2001/07/25 22:57:13 jongfoster
844 * __BEOS__ no longer overrides FEATURE_PTHREAD.
845 * This is because FEATURE_PTHREAD will soon be widely used, so I
846 * want to keep it simple.
848 * Revision 1.29 2001/07/24 12:47:06 oes
849 * Applied BeOS support update by Eugenia
851 * Revision 1.28 2001/07/23 13:26:12 oes
852 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
854 * Revision 1.27 2001/07/19 19:09:47 haroon
855 * - Added code to take care of the situation where while processing the first
856 * server response (which includes the server header), after finding the end
857 * of the headers we were not looking past the end of the headers for
858 * content modification. I enabled it for filter_popups.
859 * Someone else should look to see if other similar operations should be
860 * done to the discarded portion of the buffer.
862 * Note 2001/07/20: No, the other content modification mechanisms will process
863 * the whole iob later anyway. --oes
865 * Revision 1.26 2001/07/18 12:31:36 oes
868 * Revision 1.25 2001/07/15 19:43:49 jongfoster
869 * Supports POSIX threads.
870 * Also removed some unused #includes.
872 * Revision 1.24 2001/07/13 14:00:40 oes
873 * - Generic content modification scheme:
874 * Each feature has its own applicability flag that is set
875 * from csp->action->flags.
876 * Replaced the "filtering" int flag , by a function pointer
877 * "content_filter" to the function that will do the content
878 * modification. If it is != NULL, the document will be buffered
879 * and processed through *content_filter, which must set
880 * csp->content_length and return a modified copy of the body
881 * or return NULL (on failiure).
882 * - Changed csp->is_text to the more generic bitmap csp->content_type
883 * which can currently take the valued CT_TEXT or CT_GIF
885 * - Removed all #ifdef PCRS
887 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
888 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
889 * the `block_acl' call. This prevents a compilation error when the user
890 * does not wish to use the "ACL" feature.
892 * Revision 1.22 2001/06/29 21:45:41 oes
893 * Indentation, CRLF->LF, Tab-> Space
895 * Revision 1.21 2001/06/29 13:29:36 oes
896 * - Cleaned up, improved comments
897 * - Unified all possible interceptors (CGI,
898 * block, trust, fast_redirect) in one
899 * place, with one (CGI) answer generation
900 * mechansim. Much clearer now.
901 * - Removed the GIF image generation, which
902 * is now done in filters.c:block_url()
903 * - Made error conditions like domain lookup
904 * failiure or (various) problems while talking
905 * to the server use cgi.c:error_response()
906 * instead of generating HTML/HTTP in chat() (yuck!)
907 * - Removed logentry from cancelled commit
909 * Revision 1.20 2001/06/09 10:55:28 jongfoster
910 * Changing BUFSIZ ==> BUFFER_SIZE
912 * Revision 1.19 2001/06/07 23:12:52 jongfoster
913 * Replacing function pointer in struct gateway with a directly
914 * called function forwarded_connect().
915 * Replacing struct gateway with struct forward_spec
917 * Revision 1.18 2001/06/03 19:12:16 oes
918 * introduced new cgi handling
920 * Revision 1.17 2001/06/01 20:07:23 jongfoster
921 * Now uses action +image-blocker{} rather than config->tinygif
923 * Revision 1.16 2001/06/01 18:49:17 jongfoster
924 * Replaced "list_share" with "list" - the tiny memory gain was not
925 * worth the extra complexity.
927 * Revision 1.15 2001/05/31 21:24:47 jongfoster
928 * Changed "permission" to "action" throughout.
929 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
930 * Moved vanilla wafer check into chat(), since we must now
931 * decide whether or not to add it based on the URL.
933 * Revision 1.14 2001/05/29 20:14:01 joergs
934 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
937 * Revision 1.13 2001/05/29 09:50:24 jongfoster
938 * Unified blocklist/imagelist/permissionslist.
939 * File format is still under discussion, but the internal changes
942 * Also modified interceptor behaviour:
943 * - We now intercept all URLs beginning with one of the following
944 * prefixes (and *only* these prefixes):
946 * * http://ijbswa.sf.net/config/
947 * * http://ijbswa.sourceforge.net/config/
948 * - New interceptors "home page" - go to http://i.j.b/ to see it.
949 * - Internal changes so that intercepted and fast redirect pages
950 * are not replaced with an image.
951 * - Interceptors now have the option to send a binary page direct
952 * to the client. (i.e. ijb-send-banner uses this)
953 * - Implemented show-url-info interceptor. (Which is why I needed
954 * the above interceptors changes - a typical URL is
955 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
956 * The previous mechanism would not have intercepted that, and
957 * if it had been intercepted then it then it would have replaced
960 * Revision 1.12 2001/05/27 22:17:04 oes
962 * - re_process_buffer no longer writes the modified buffer
963 * to the client, which was very ugly. It now returns the
964 * buffer, which it is then written by chat.
966 * - content_length now adjusts the Content-Length: header
967 * for modified documents rather than crunch()ing it.
968 * (Length info in csp->content_length, which is 0 for
969 * unmodified documents)
971 * - For this to work, sed() is called twice when filtering.
973 * Revision 1.11 2001/05/26 17:27:53 jongfoster
974 * Added support for CLF and fixed LOG_LEVEL_LOG.
975 * Also did CRLF->LF fix of my previous patch.
977 * Revision 1.10 2001/05/26 15:26:15 jongfoster
978 * ACL feature now provides more security by immediately dropping
979 * connections from untrusted hosts.
981 * Revision 1.9 2001/05/26 00:28:36 jongfoster
982 * Automatic reloading of config file.
983 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
984 * Most of the global variables have been moved to a new
985 * struct configuration_spec, accessed through csp->config->globalname
986 * Most of the globals remaining are used by the Win32 GUI.
988 * Revision 1.8 2001/05/25 22:43:18 jongfoster
989 * Fixing minor memory leak and buffer overflow.
991 * Revision 1.7 2001/05/25 22:34:30 jongfoster
994 * Revision 1.6 2001/05/23 00:13:58 joergs
995 * AmigaOS support fixed.
997 * Revision 1.5 2001/05/22 18:46:04 oes
999 * - Enabled filtering banners by size rather than URL
1000 * by adding patterns that replace all standard banner
1001 * sizes with the "Junkbuster" gif to the re_filterfile
1003 * - Enabled filtering WebBugs by providing a pattern
1004 * which kills all 1x1 images
1006 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
1007 * which is selected by the (nonstandard and therefore
1008 * capital) letter 'U' in the option string.
1009 * It causes the quantifiers to be ungreedy by default.
1010 * Appending a ? turns back to greedy (!).
1012 * - Added a new interceptor ijb-send-banner, which
1013 * sends back the "Junkbuster" gif. Without imagelist or
1014 * MSIE detection support, or if tinygif = 1, or the
1015 * URL isn't recognized as an imageurl, a lame HTML
1016 * explanation is sent instead.
1018 * - Added new feature, which permits blocking remote
1019 * script redirects and firing back a local redirect
1021 * The feature is conditionally compiled, i.e. it
1022 * can be disabled with --disable-fast-redirects,
1023 * plus it must be activated by a "fast-redirects"
1024 * line in the config file, has its own log level
1025 * and of course wants to be displayed by show-proxy-args
1026 * Note: Boy, all the #ifdefs in 1001 locations and
1027 * all the fumbling with configure.in and acconfig.h
1028 * were *way* more work than the feature itself :-(
1030 * - Because a generic redirect template was needed for
1031 * this, tinygif = 3 now uses the same.
1033 * - Moved GIFs, and other static HTTP response templates
1036 * - Some minor fixes
1038 * - Removed some >400 CRs again (Jon, you really worked
1041 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1042 * Made failure to bind() a fatal error.
1044 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1045 * Version 2.9.4 checkin.
1046 * - Merged popupfile and cookiefile, and added control over PCRS
1047 * filtering, in new "permissionsfile".
1048 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1049 * file error you now get a message box (in the Win32 GUI) rather
1050 * than the program exiting with no explanation.
1051 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1053 * - Removed tabs from "config"
1054 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1055 * - Bumped up version number.
1057 * Revision 1.2 2001/05/17 22:34:44 oes
1058 * - Added hint on GIF char array generation to jcc.c
1059 * - Cleaned CRLF's from the sources and related files
1060 * - Repaired logging for REF and FRC
1062 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1063 * Initial import of version 2.9.3 source tree
1066 *********************************************************************/
1072 #include <sys/types.h>
1081 # ifndef FEATURE_PTHREAD
1085 # include <windows.h>
1086 # include <process.h>
1087 # endif /* ndef FEATURE_PTHREAD */
1090 # ifndef _WIN_CONSOLE
1091 # include "w32log.h"
1092 # endif /* ndef _WIN_CONSOLE */
1093 # include "w32svrapi.h"
1095 #else /* ifndef _WIN32 */
1097 # if !defined (__OS2__)
1098 # include <unistd.h>
1099 # include <sys/wait.h>
1100 # endif /* ndef __OS2__ */
1101 # include <sys/time.h>
1102 # include <sys/stat.h>
1103 # include <sys/ioctl.h>
1106 #include <sys/termios.h>
1114 # include <signal.h>
1117 # include <socket.h> /* BeOS has select() for sockets only. */
1118 # include <OS.h> /* declarations for threads and stuff. */
1121 # if defined(__EMX__) || defined(__OS2__)
1122 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1127 #define bzero(B,N) memset(B,0x00,n)
1131 # include <select.h>
1136 #include "project.h"
1139 #include "filters.h"
1140 #include "loaders.h"
1141 #include "parsers.h"
1142 #include "miscutil.h"
1144 #include "jbsockets.h"
1145 #include "gateway.h"
1146 #include "actions.h"
1148 #include "loadcfg.h"
1149 #include "urlmatch.h"
1151 const char jcc_h_rcs[] = JCC_H_VERSION;
1152 const char project_h_rcs[] = PROJECT_H_VERSION;
1155 struct client_state clients[1];
1156 struct file_list files[1];
1158 #ifdef FEATURE_STATISTICS
1159 int urls_read = 0; /* total nr of urls read inc rejected */
1160 int urls_rejected = 0; /* total nr of urls rejected */
1161 #endif /* def FEATURE_STATISTICS */
1163 #ifdef FEATURE_GRACEFUL_TERMINATION
1164 int g_terminate = 0;
1167 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1168 static void sig_handler(int the_signal);
1170 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1171 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1172 static jb_err get_server_headers(struct client_state *csp);
1173 static const char *crunch_reason(const struct http_response *rsp);
1174 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1176 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1178 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1179 static jb_err change_request_destination(struct client_state *csp);
1180 static void chat(struct client_state *csp);
1181 static void serve(struct client_state *csp);
1182 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1183 static void usage(const char *myname);
1185 static void initialize_mutexes(void);
1186 static jb_socket bind_port_helper(struct configuration_spec *config);
1187 static void listen_loop(void);
1190 void serve(struct client_state *csp);
1191 #else /* ifndef AMIGA */
1192 static void serve(struct client_state *csp);
1193 #endif /* def AMIGA */
1196 static int32 server_thread(void *data);
1197 #endif /* def __BEOS__ */
1200 #define sleep(N) Sleep(((N) * 1000))
1204 #define sleep(N) DosSleep(((N) * 100))
1207 #ifdef MUTEX_LOCKS_AVAILABLE
1209 * XXX: Does the locking stuff really belong in this file?
1211 privoxy_mutex_t log_mutex;
1212 privoxy_mutex_t log_init_mutex;
1213 privoxy_mutex_t connection_reuse_mutex;
1215 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1216 privoxy_mutex_t resolver_mutex;
1217 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1219 #ifndef HAVE_GMTIME_R
1220 privoxy_mutex_t gmtime_mutex;
1221 #endif /* ndef HAVE_GMTIME_R */
1223 #ifndef HAVE_LOCALTIME_R
1224 privoxy_mutex_t localtime_mutex;
1225 #endif /* ndef HAVE_GMTIME_R */
1228 privoxy_mutex_t rand_mutex;
1229 #endif /* ndef HAVE_RANDOM */
1231 #endif /* def MUTEX_LOCKS_AVAILABLE */
1234 const char *basedir = NULL;
1235 const char *pidfile = NULL;
1236 static int received_hup_signal = 0;
1237 #endif /* defined unix */
1239 /* HTTP snipplets. */
1240 static const char CSUCCEED[] =
1241 "HTTP/1.0 200 Connection established\r\n"
1242 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1244 static const char CHEADER[] =
1245 "HTTP/1.0 400 Invalid header received from client\r\n"
1246 "Proxy-Agent: Privoxy " VERSION "\r\n"
1247 "Content-Type: text/plain\r\n"
1248 "Connection: close\r\n\r\n"
1249 "Invalid header received from client.\r\n";
1251 static const char FTP_RESPONSE[] =
1252 "HTTP/1.0 400 Invalid request received from client\r\n"
1253 "Content-Type: text/plain\r\n"
1254 "Connection: close\r\n\r\n"
1255 "Invalid request. Privoxy doesn't support FTP.\r\n";
1257 static const char GOPHER_RESPONSE[] =
1258 "HTTP/1.0 400 Invalid request received from client\r\n"
1259 "Content-Type: text/plain\r\n"
1260 "Connection: close\r\n\r\n"
1261 "Invalid request. Privoxy doesn't support gopher.\r\n";
1263 /* XXX: should be a template */
1264 static const char MISSING_DESTINATION_RESPONSE[] =
1265 "HTTP/1.0 400 Bad request received from client\r\n"
1266 "Proxy-Agent: Privoxy " VERSION "\r\n"
1267 "Content-Type: text/plain\r\n"
1268 "Connection: close\r\n\r\n"
1269 "Bad request. Privoxy was unable to extract the destination.\r\n";
1271 /* XXX: should be a template */
1272 static const char NO_SERVER_DATA_RESPONSE[] =
1273 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1274 "Proxy-Agent: Privoxy " VERSION "\r\n"
1275 "Content-Type: text/plain\r\n"
1276 "Connection: close\r\n\r\n"
1277 "Empty server or forwarder response.\r\n"
1278 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1280 /* XXX: should be a template */
1281 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1282 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1283 "Proxy-Agent: Privoxy " VERSION "\r\n"
1284 "Content-Type: text/plain\r\n"
1285 "Connection: close\r\n\r\n"
1286 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1289 /* XXX: should be a template */
1290 static const char NULL_BYTE_RESPONSE[] =
1291 "HTTP/1.0 400 Bad request received from client\r\n"
1292 "Proxy-Agent: Privoxy " VERSION "\r\n"
1293 "Content-Type: text/plain\r\n"
1294 "Connection: close\r\n\r\n"
1295 "Bad request. Null byte(s) before end of request.\r\n";
1298 /* XXX: should be a template */
1299 static const char MESSED_UP_REQUEST_RESPONSE[] =
1300 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1301 "Proxy-Agent: Privoxy " VERSION "\r\n"
1302 "Content-Type: text/plain\r\n"
1303 "Connection: close\r\n\r\n"
1304 "Bad request. Messed up with header filters.\r\n";
1306 /* A function to crunch a response */
1307 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1309 /* Crunch function flags */
1310 #define CF_NO_FLAGS 0
1311 /* Cruncher applies to forced requests as well */
1312 #define CF_IGNORE_FORCE 1
1313 /* Crunched requests are counted for the block statistics */
1314 #define CF_COUNT_AS_REJECT 2
1316 /* A crunch function and its flags */
1319 const crunch_func_ptr cruncher;
1323 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1325 /* Complete list of cruncher functions */
1326 static const struct cruncher crunchers_all[] = {
1327 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1328 { block_url, CF_COUNT_AS_REJECT },
1329 #ifdef FEATURE_TRUST
1330 { trust_url, CF_COUNT_AS_REJECT },
1331 #endif /* def FEATURE_TRUST */
1332 { redirect_url, CF_NO_FLAGS },
1333 { dispatch_cgi, CF_IGNORE_FORCE},
1337 /* Light version, used after tags are applied */
1338 static const struct cruncher crunchers_light[] = {
1339 { block_url, CF_COUNT_AS_REJECT },
1340 { redirect_url, CF_NO_FLAGS },
1346 * XXX: Don't we really mean
1352 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1353 /*********************************************************************
1355 * Function : sig_handler
1357 * Description : Signal handler for different signals.
1358 * Exit gracefully on TERM and INT
1359 * or set a flag that will cause the errlog
1360 * to be reopened by the main thread on HUP.
1363 * 1 : the_signal = the signal cause this function to call
1367 *********************************************************************/
1368 static void sig_handler(int the_signal)
1374 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1386 received_hup_signal = 1;
1392 * We shouldn't be here, unless we catch signals
1393 * in main() that we can't handle here!
1395 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1403 /*********************************************************************
1405 * Function : client_protocol_is_unsupported
1407 * Description : Checks if the client used a known unsupported
1408 * protocol and deals with it by sending an error
1412 * 1 : csp = Current client state (buffers, headers, etc...)
1413 * 2 : req = the first request line send by the client
1415 * Returns : TRUE if an error response has been generated, or
1416 * FALSE if the request doesn't look invalid.
1418 *********************************************************************/
1419 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1422 * If it's a FTP or gopher request, we don't support it.
1424 * These checks are better than nothing, but they might
1425 * not work in all configurations and some clients might
1426 * have problems digesting the answer.
1428 * They should, however, never cause more problems than
1429 * Privoxy's old behaviour (returning the misleading HTML
1432 * "Could not resolve http://(ftp|gopher)://example.org").
1434 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1436 const char *response = NULL;
1437 const char *protocol = NULL;
1439 if (!strncmpic(req, "GET ftp://", 10))
1441 response = FTP_RESPONSE;
1446 response = GOPHER_RESPONSE;
1447 protocol = "GOPHER";
1449 log_error(LOG_LEVEL_ERROR,
1450 "%s tried to use Privoxy as %s proxy: %s",
1451 csp->ip_addr_str, protocol, req);
1452 log_error(LOG_LEVEL_CLF,
1453 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1455 write_socket(csp->cfd, response, strlen(response));
1464 /*********************************************************************
1466 * Function : get_request_destination_elsewhere
1468 * Description : If the client's request was redirected into
1469 * Privoxy without the client's knowledge,
1470 * the request line lacks the destination host.
1472 * This function tries to get it elsewhere,
1473 * provided accept-intercepted-requests is enabled.
1475 * "Elsewhere" currently only means "Host: header",
1476 * but in the future we may ask the redirecting
1477 * packet filter to look the destination up.
1479 * If the destination stays unknown, an error
1480 * response is send to the client and headers
1481 * are freed so that chat() can return directly.
1484 * 1 : csp = Current client state (buffers, headers, etc...)
1485 * 2 : headers = a header list
1487 * Returns : JB_ERR_OK if the destination is now known, or
1488 * JB_ERR_PARSE if it isn't.
1490 *********************************************************************/
1491 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1495 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1497 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1498 " Privoxy isn't configured to accept intercepted requests.",
1499 csp->ip_addr_str, csp->http->cmd);
1500 /* XXX: Use correct size */
1501 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1502 csp->ip_addr_str, csp->http->cmd);
1504 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1505 destroy_list(headers);
1507 return JB_ERR_PARSE;
1509 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1511 /* Split the domain we just got for pattern matching */
1512 init_domain_components(csp->http);
1518 /* We can't work without destination. Go spread the news.*/
1520 req = list_to_text(headers);
1522 /* XXX: Use correct size */
1523 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1524 csp->ip_addr_str, csp->http->cmd);
1525 log_error(LOG_LEVEL_ERROR,
1526 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1527 csp->ip_addr_str, csp->http->cmd, req);
1530 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1531 destroy_list(headers);
1533 return JB_ERR_PARSE;
1536 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1537 * to get the destination IP address, use it as host directly
1538 * or do a reverse DNS lookup first.
1543 /*********************************************************************
1545 * Function : get_server_headers
1547 * Description : Parses server headers in iob and fills them
1548 * into csp->headers so that they can later be
1552 * 1 : csp = Current client state (buffers, headers, etc...)
1554 * Returns : JB_ERR_OK if everything went fine, or
1555 * JB_ERR_PARSE if the headers were incomplete.
1557 *********************************************************************/
1558 static jb_err get_server_headers(struct client_state *csp)
1560 int continue_hack_in_da_house = 0;
1563 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1568 * continue hack in da house. Ignore the ending of
1569 * this head and continue enlisting header lines.
1570 * The reason is described below.
1572 enlist(csp->headers, "");
1573 continue_hack_in_da_house = 0;
1576 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1579 * It's a bodyless continue response, don't
1580 * stop header parsing after reaching its end.
1582 * As a result Privoxy will concatenate the
1583 * next response's head and parse and deliver
1584 * the headers as if they belonged to one request.
1586 * The client will separate them because of the
1587 * empty line between them.
1589 * XXX: What we're doing here is clearly against
1590 * the intended purpose of the continue header,
1591 * and under some conditions (HTTP/1.0 client request)
1592 * it's a standard violation.
1594 * Anyway, "sort of against the spec" is preferable
1595 * to "always getting confused by Continue responses"
1596 * (Privoxy's behaviour before this hack was added)
1598 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1599 continue_hack_in_da_house = 1;
1601 else if (*header == '\0')
1604 * If the header is empty, but the Continue hack
1605 * isn't active, we can assume that we reached the
1606 * end of the buffer before we hit the end of the
1609 * Inform the caller an let it decide how to handle it.
1611 return JB_ERR_PARSE;
1614 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1617 * XXX: Should we quit the request and return a
1618 * out of memory error page instead?
1620 log_error(LOG_LEVEL_ERROR,
1621 "Out of memory while enlisting server headers. %s lost.",
1631 /*********************************************************************
1633 * Function : crunch_reason
1635 * Description : Translates the crunch reason code into a string.
1638 * 1 : rsp = a http_response
1640 * Returns : A string with the crunch reason or an error description.
1642 *********************************************************************/
1643 static const char *crunch_reason(const struct http_response *rsp)
1645 char * reason = NULL;
1647 assert(rsp != NULL);
1650 return "Internal error while searching for crunch reason";
1653 switch (rsp->reason)
1655 case RSP_REASON_UNSUPPORTED:
1656 reason = "Unsupported HTTP feature";
1658 case RSP_REASON_BLOCKED:
1661 case RSP_REASON_UNTRUSTED:
1662 reason = "Untrusted";
1664 case RSP_REASON_REDIRECTED:
1665 reason = "Redirected";
1667 case RSP_REASON_CGI_CALL:
1668 reason = "CGI Call";
1670 case RSP_REASON_NO_SUCH_DOMAIN:
1671 reason = "DNS failure";
1673 case RSP_REASON_FORWARDING_FAILED:
1674 reason = "Forwarding failed";
1676 case RSP_REASON_CONNECT_FAILED:
1677 reason = "Connection failure";
1679 case RSP_REASON_OUT_OF_MEMORY:
1680 reason = "Out of memory (may mask other reasons)";
1683 reason = "No reason recorded";
1691 /*********************************************************************
1693 * Function : send_crunch_response
1695 * Description : Delivers already prepared response for
1696 * intercepted requests, logs the interception
1697 * and frees the response.
1700 * 1 : csp = Current client state (buffers, headers, etc...)
1701 * 1 : rsp = Fully prepared response. Will be freed on exit.
1703 * Returns : Nothing.
1705 *********************************************************************/
1706 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1708 const struct http_request *http = csp->http;
1709 char status_code[4];
1711 assert(rsp != NULL);
1712 assert(rsp->head != NULL);
1717 * Not supposed to happen. If it does
1718 * anyway, treat it as an unknown error.
1720 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1721 /* return code doesn't matter */
1726 /* If rsp is still NULL, we have serious internal problems. */
1727 log_error(LOG_LEVEL_FATAL,
1728 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1732 * Extract the status code from the actual head
1733 * that was send to the client. It is the only
1734 * way to get it right for all requests, including
1735 * the fixed ones for out-of-memory problems.
1737 * A head starts like this: 'HTTP/1.1 200...'
1741 status_code[0] = rsp->head[9];
1742 status_code[1] = rsp->head[10];
1743 status_code[2] = rsp->head[11];
1744 status_code[3] = '\0';
1746 /* Write the answer to the client */
1747 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1748 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1750 /* There is nothing we can do about it. */
1751 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1754 /* Log that the request was crunched and why. */
1755 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1756 http->hostport, http->path, crunch_reason(rsp));
1757 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1758 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1760 /* Clean up and return */
1761 if (cgi_error_memory() != rsp)
1763 free_http_response(rsp);
1770 /*********************************************************************
1772 * Function : request_contains_null_bytes
1774 * Description : Checks for NULL bytes in the request and sends
1775 * an error message to the client if any were found.
1777 * XXX: currently not used, see comment in chat().
1780 * 1 : csp = Current client state (buffers, headers, etc...)
1781 * 2 : buf = Data from the client's request to check.
1782 * 3 : len = The data length.
1784 * Returns : TRUE if the request contained one or more NULL bytes, or
1787 *********************************************************************/
1788 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1790 size_t c_len; /* Request lenght when treated as C string */
1792 c_len = strlen(buf);
1797 * Null byte(s) found. Log the request,
1798 * return an error response and hang up.
1800 size_t tmp_len = c_len;
1805 * Replace NULL byte(s) with '°' characters
1806 * so the request can be logged as string.
1807 * XXX: Is there a better replacement character?
1810 tmp_len += strlen(buf+tmp_len);
1811 } while (tmp_len < len);
1813 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1814 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1815 log_error(LOG_LEVEL_HEADER,
1816 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1818 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1820 /* XXX: Log correct size */
1821 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1831 /*********************************************************************
1833 * Function : crunch_response_triggered
1835 * Description : Checks if the request has to be crunched,
1836 * and delivers the crunch response if necessary.
1839 * 1 : csp = Current client state (buffers, headers, etc...)
1840 * 2 : crunchers = list of cruncher functions to run
1842 * Returns : TRUE if the request was answered with a crunch response
1845 *********************************************************************/
1846 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1848 struct http_response *rsp = NULL;
1849 const struct cruncher *c;
1852 * If CGI request crunching is disabled,
1853 * check the CGI dispatcher out of order to
1854 * prevent unintentional blocks or redirects.
1856 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1857 && (NULL != (rsp = dispatch_cgi(csp))))
1859 /* Deliver, log and free the interception response. */
1860 send_crunch_response(csp, rsp);
1864 for (c = crunchers; c->cruncher != NULL; c++)
1867 * Check the cruncher if either Privoxy is toggled
1868 * on and the request isn't forced, or if the cruncher
1869 * applies to forced requests as well.
1871 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1872 !(csp->flags & CSP_FLAG_FORCED)) ||
1873 (c->flags & CF_IGNORE_FORCE))
1875 rsp = c->cruncher(csp);
1878 /* Deliver, log and free the interception response. */
1879 send_crunch_response(csp, rsp);
1880 #ifdef FEATURE_STATISTICS
1881 if (c->flags & CF_COUNT_AS_REJECT)
1883 csp->flags |= CSP_FLAG_REJECTED;
1885 #endif /* def FEATURE_STATISTICS */
1896 /*********************************************************************
1898 * Function : build_request_line
1900 * Description : Builds the HTTP request line.
1902 * If a HTTP forwarder is used it expects the whole URL,
1903 * web servers only get the path.
1906 * 1 : csp = Current client state (buffers, headers, etc...)
1907 * 2 : fwd = The forwarding spec used for the request
1908 * XXX: Should use http->fwd instead.
1909 * 3 : request_line = The old request line which will be replaced.
1911 * Returns : Nothing. Terminates in case of memory problems.
1913 *********************************************************************/
1914 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1916 struct http_request *http = csp->http;
1918 assert(http->ssl == 0);
1921 * Downgrade http version from 1.1 to 1.0
1922 * if +downgrade action applies.
1924 if ( (csp->action->flags & ACTION_DOWNGRADE)
1925 && (!strcmpic(http->ver, "HTTP/1.1")))
1928 http->ver = strdup("HTTP/1.0");
1930 if (http->ver == NULL)
1932 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1937 * Rebuild the request line.
1939 freez(*request_line);
1940 *request_line = strdup(http->gpc);
1941 string_append(request_line, " ");
1943 if (fwd->forward_host)
1945 string_append(request_line, http->url);
1949 string_append(request_line, http->path);
1951 string_append(request_line, " ");
1952 string_append(request_line, http->ver);
1954 if (*request_line == NULL)
1956 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1958 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1962 /*********************************************************************
1964 * Function : change_request_destination
1966 * Description : Parse a (rewritten) request line and regenerate
1967 * the http request data.
1970 * 1 : csp = Current client state (buffers, headers, etc...)
1972 * Returns : Forwards the parse_http_request() return code.
1973 * Terminates in case of memory problems.
1975 *********************************************************************/
1976 static jb_err change_request_destination(struct client_state *csp)
1978 struct http_request *http = csp->http;
1981 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1982 free_http_request(http);
1983 err = parse_http_request(csp->headers->first->str, http, csp);
1984 if (JB_ERR_OK != err)
1986 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1987 jb_err_to_string(err));
1991 /* XXX: ocmd is a misleading name */
1992 http->ocmd = strdup(http->cmd);
1993 if (http->ocmd == NULL)
1995 log_error(LOG_LEVEL_FATAL,
1996 "Out of memory copying rewritten HTTP request line");
2004 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2005 /*********************************************************************
2007 * Function : server_response_is_complete
2009 * Description : Determines whether we should stop reading
2010 * from the server socket.
2013 * 1 : csp = Current client state (buffers, headers, etc...)
2014 * 2 : content_length = Length of content received so far.
2016 * Returns : TRUE if the response is complete,
2019 *********************************************************************/
2020 static int server_response_is_complete(struct client_state *csp, size_t content_length)
2022 int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2024 if (!strcmpic(csp->http->gpc, "HEAD"))
2027 * "HEAD" implies no body, we are thus expecting
2028 * no content. XXX: incomplete "list" of methods?
2030 log_error(LOG_LEVEL_INFO, "Method %s implies no body.", csp->http->gpc);
2031 csp->expected_content_length = 0;
2032 content_length_known = TRUE;
2035 if (csp->http->status == 304)
2038 * Expect no body. XXX: incomplete "list" of status codes?
2040 log_error(LOG_LEVEL_INFO, "Status code %d implies no body.", csp->http->status);
2041 csp->expected_content_length = 0;
2042 content_length_known = TRUE;
2045 return (content_length_known && ((0 == csp->expected_content_length)
2046 || (csp->expected_content_length <= content_length)));
2048 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2051 /*********************************************************************
2055 * Description : Once a connection to the client has been accepted,
2056 * this function is called (via serve()) to handle the
2057 * main business of the communication. When this
2058 * function returns, the caller must close the client
2061 * FIXME: chat is nearly thousand lines long.
2065 * 1 : csp = Current client state (buffers, headers, etc...)
2067 * Returns : Nothing.
2069 *********************************************************************/
2070 static void chat(struct client_state *csp)
2072 char buf[BUFFER_SIZE];
2080 int ms_iis5_hack = 0;
2081 size_t byte_count = 0;
2082 int forwarded_connect_retries = 0;
2083 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2084 const struct forward_spec * fwd;
2085 struct http_request *http;
2086 int len; /* for buffer sizes (and negative error codes) */
2089 /* Function that does the content filtering for the current request */
2090 filter_function_ptr content_filter = NULL;
2092 /* Skeleton for HTTP response, if we should intercept the request */
2093 struct http_response *rsp;
2095 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2096 struct list header_list;
2097 struct list *headers = &header_list;
2101 memset(buf, 0, sizeof(buf));
2104 * Read the client's request. Note that since we're not using select() we
2105 * could get blocked here if a client connected, then didn't say anything!
2110 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2112 if (len <= 0) break; /* error! */
2115 * If there is no memory left for buffering the
2116 * request, there is nothing we can do but hang up
2118 if (add_to_iob(csp, buf, len))
2123 req = get_header(csp->iob);
2125 } while ((NULL != req) && ('\0' == *req));
2127 if ((NULL != req) && ('\0' != *req))
2129 /* Request received. Validate and parse it. */
2133 * XXX: Temporary disabled to prevent problems
2134 * with POST requests whose bodies are allowed to
2135 * contain NULL bytes. BR#1730105.
2137 * The main purpose of this check is to properly
2138 * log stuff like BitTorrent traffic and other junk
2139 * that hits public proxies. It's not required for
2140 * Privoxy to functions as those requests are discarded
2143 * It probably should be rewritten to only check
2144 * the head of the request. Another option would
2145 * be to let all POST requests pass, although that
2146 * may not be good enough.
2148 if (request_contains_null_bytes(csp, buf, len))
2150 /* NULL bytes found and dealt with, just hang up. */
2155 /* Does the request line look invalid? */
2156 if (client_protocol_is_unsupported(csp, req))
2159 * Yes. The request has already been
2160 * answered with a error response, the buffers
2161 * were freed and we're done with chatting.
2166 #ifdef FEATURE_FORCE_LOAD
2168 * If this request contains the FORCE_PREFIX and blocks
2169 * aren't enforced, get rid of it and set the force flag.
2171 if (strstr(req, FORCE_PREFIX))
2173 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2175 log_error(LOG_LEVEL_FORCE,
2176 "Ignored force prefix in request: \"%s\".", req);
2180 strclean(req, FORCE_PREFIX);
2181 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2182 csp->flags |= CSP_FLAG_FORCED;
2186 #endif /* def FEATURE_FORCE_LOAD */
2187 err = parse_http_request(req, http, csp);
2188 if (JB_ERR_OK != err)
2190 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2196 if (http->cmd == NULL)
2198 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2199 /* XXX: Use correct size */
2200 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2201 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2203 free_http_request(http);
2207 /* grab the rest of the client's headers */
2211 p = get_header(csp->iob);
2215 /* There are no additional headers to read. */
2222 * We didn't receive a complete header
2223 * line yet, get the rest of it.
2225 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2228 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2229 destroy_list(headers);
2233 if (add_to_iob(csp, buf, len))
2236 * If there is no memory left for buffering the
2237 * request, there is nothing we can do but hang up
2239 destroy_list(headers);
2246 * We were able to read a complete
2247 * header and can finaly enlist it.
2254 if (http->host == NULL)
2257 * If we still don't know the request destination,
2258 * the request is invalid or the client uses
2259 * Privoxy without its knowledge.
2261 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2264 * Our attempts to get the request destination
2265 * elsewhere failed or Privoxy is configured
2266 * to only accept proxy requests.
2268 * An error response has already been send
2269 * and we're done here.
2276 * Determine the actions for this URL
2278 #ifdef FEATURE_TOGGLE
2279 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2281 /* Most compatible set of actions (i.e. none) */
2282 init_current_action(csp->action);
2285 #endif /* ndef FEATURE_TOGGLE */
2287 get_url_actions(csp, http);
2291 * Save a copy of the original request for logging
2293 http->ocmd = strdup(http->cmd);
2295 if (http->ocmd == NULL)
2297 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2300 enlist(csp->headers, http->cmd);
2302 /* Append the previously read headers */
2303 list_append_list_unique(csp->headers, headers);
2304 destroy_list(headers);
2306 err = sed(csp, FILTER_CLIENT_HEADERS);
2307 if (JB_ERR_OK != err)
2309 assert(err == JB_ERR_PARSE);
2310 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2312 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2314 /* Check request line for rewrites. */
2315 if ((NULL == csp->headers->first->str)
2316 || (strcmp(http->cmd, csp->headers->first->str) &&
2317 (JB_ERR_OK != change_request_destination(csp))))
2320 * A header filter broke the request line - bail out.
2322 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2323 /* XXX: Use correct size */
2324 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2325 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2327 free_http_request(http);
2331 /* decide how to route the HTTP request */
2332 fwd = forward_url(csp, http);
2335 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2336 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2340 * build the http request to send to the server
2341 * we have to do one of the following:
2343 * create = use the original HTTP request to create a new
2344 * HTTP request that has either the path component
2345 * without the http://domainspec (w/path) or the
2346 * full orininal URL (w/url)
2347 * Note that the path and/or the HTTP version may
2348 * have been altered by now.
2350 * connect = Open a socket to the host:port of the server
2351 * and short-circuit server and client socket.
2353 * pass = Pass the request unchanged if forwarding a CONNECT
2354 * request to a parent proxy. Note that we'll be sending
2355 * the CFAIL message ourselves if connecting to the parent
2356 * fails, but we won't send a CSUCCEED message if it works,
2357 * since that would result in a double message (ours and the
2358 * parent's). After sending the request to the parent, we simply
2361 * here's the matrix:
2364 * +--------+--------+
2366 * 0 | create | connect|
2368 * Forwarding +--------+--------+
2370 * 1 | create | pass |
2372 * +--------+--------+
2376 if (http->ssl && connect_port_is_forbidden(csp))
2378 const char *acceptable_connect_ports =
2379 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2380 assert(NULL != acceptable_connect_ports);
2381 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2382 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2383 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2384 csp->action->flags |= ACTION_BLOCK;
2390 freez(csp->headers->first->str);
2391 build_request_line(csp, fwd, &csp->headers->first->str);
2394 hdr = list_to_text(csp->headers);
2397 /* FIXME Should handle error properly */
2398 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2402 * We have a request. Check if one of the crunchers wants it.
2404 if (crunch_response_triggered(csp, crunchers_all))
2407 * Yes. The client got the crunch response
2408 * and we are done here after cleaning up.
2411 list_remove_all(csp->headers);
2417 * The headers can't be removed earlier because
2418 * they were still needed for the referrer check
2419 * in case of CGI crunches.
2421 * XXX: Would it be worth to move the referrer check
2422 * into client_referrer() and set a flag if it's trusted?
2424 list_remove_all(csp->headers);
2426 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2428 if (fwd->forward_host)
2430 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2431 fwd->forward_host, fwd->forward_port, http->hostport);
2435 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2438 /* here we connect to the server, gateway, or the forwarder */
2440 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2441 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2443 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2444 forwarded_connect_retries, http->hostport);
2447 if (csp->sfd == JB_INVALID_SOCKET)
2449 if (fwd->type != SOCKS_NONE)
2452 rsp = error_response(csp, "forwarding-failed", errno);
2454 else if (errno == EINVAL)
2456 rsp = error_response(csp, "no-such-domain", errno);
2460 rsp = error_response(csp, "connect-failed", errno);
2461 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2465 /* Write the answer to the client */
2468 send_crunch_response(csp, rsp);
2475 if (fwd->forward_host || (http->ssl == 0))
2478 * Write the client's (modified) header to the server
2479 * (along with anything else that may be in the buffer)
2481 if (write_socket(csp->sfd, hdr, strlen(hdr))
2482 || (flush_socket(csp->sfd, csp->iob) < 0))
2484 log_error(LOG_LEVEL_CONNECT,
2485 "write header to: %s failed: %E", http->hostport);
2487 rsp = error_response(csp, "connect-failed", errno);
2490 send_crunch_response(csp, rsp);
2500 * We're running an SSL tunnel and we're not forwarding,
2501 * so just send the "connect succeeded" message to the
2502 * client, flush the rest, and get out of the way.
2504 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2512 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2514 /* we're finished with the client's header */
2517 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2519 /* pass data between the client and server
2520 * until one or the other shuts down the connection.
2529 * FD_ZERO here seems to point to an errant macro which crashes.
2530 * So do this by hand for now...
2532 memset(&rfds,0x00,sizeof(fd_set));
2536 FD_SET(csp->cfd, &rfds);
2537 FD_SET(csp->sfd, &rfds);
2539 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2540 if (server_body && server_response_is_complete(csp, byte_count))
2542 log_error(LOG_LEVEL_CONNECT,
2543 "Done reading from server. Expected content length: %d. "
2544 "Actual content length: %d. Most recently received: %d.",
2545 csp->expected_content_length, byte_count, len);
2548 * XXX: should not jump around,
2549 * chat() is complicated enough already.
2553 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2555 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2559 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2564 * This is the body of the browser's request,
2565 * just read and write it.
2567 if (FD_ISSET(csp->cfd, &rfds))
2569 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2573 break; /* "game over, man" */
2576 if (write_socket(csp->sfd, buf, (size_t)len))
2578 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2585 * The server wants to talk. It could be the header or the body.
2586 * If `hdr' is null, then it's the header otherwise it's the body.
2587 * FIXME: Does `hdr' really mean `host'? No.
2589 if (FD_ISSET(csp->sfd, &rfds))
2592 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2596 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2598 if (http->ssl && (fwd->forward_host == NULL))
2601 * Just hang up. We already confirmed the client's CONNECT
2602 * request with status code 200 and unencrypted content is
2603 * no longer welcome.
2605 log_error(LOG_LEVEL_ERROR,
2606 "CONNECT already confirmed. Unable to tell the client about the problem.");
2609 else if (byte_count)
2612 * Just hang up. We already transmitted the original headers
2613 * and parts of the original content and therefore missed the
2614 * chance to send an error message (without risking data corruption).
2616 * XXX: we could retry with a fancy range request here.
2618 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2619 "Unable to tell the client about the problem.");
2623 rsp = error_response(csp, "connect-failed", errno);
2626 send_crunch_response(csp, rsp);
2632 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2633 if (csp->flags & CSP_FLAG_CHUNKED)
2635 if ((len > 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
2637 /* XXX: this is a temporary hack */
2638 log_error(LOG_LEVEL_CONNECT,
2639 "Looks like we reached the end of the last chunk: "
2640 "%d %d %d %d %d. We better stop reading.",
2641 buf[len-5], buf[len-4], buf[len-3], buf[len-2], buf[len-1]);
2642 csp->expected_content_length = byte_count + len;
2643 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2647 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2650 * Add a trailing zero to let be able to use string operations.
2651 * XXX: do we still need this with filter_popups gone?
2656 * Normally, this would indicate that we've read
2657 * as much as the server has sent us and we can
2658 * close the client connection. However, Microsoft
2659 * in its wisdom has released IIS/5 with a bug that
2660 * prevents it from sending the trailing \r\n in
2661 * a 302 redirect header (and possibly other headers).
2662 * To work around this if we've haven't parsed
2663 * a full header we'll append a trailing \r\n
2664 * and see if this now generates a valid one.
2666 * This hack shouldn't have any impacts. If we've
2667 * already transmitted the header or if this is a
2668 * SSL connection, then we won't bother with this
2669 * hack. So we only work on partially received
2670 * headers. If we append a \r\n and this still
2671 * doesn't generate a valid header, then we won't
2672 * transmit anything to the client.
2677 if (server_body || http->ssl)
2680 * If we have been buffering up the document,
2681 * now is the time to apply content modification
2682 * and send the result to the client.
2686 p = execute_content_filter(csp, content_filter);
2688 * If the content filter fails, use the original
2689 * buffer and length.
2690 * (see p != NULL ? p : csp->iob->cur below)
2694 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2697 if (JB_ERR_OK != update_server_headers(csp))
2699 log_error(LOG_LEVEL_FATAL,
2700 "Failed to update server headers. after filtering.");
2703 hdr = list_to_text(csp->headers);
2706 /* FIXME Should handle error properly */
2707 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2710 if (write_socket(csp->cfd, hdr, strlen(hdr))
2711 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2713 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2723 break; /* "game over, man" */
2727 * This is NOT the body, so
2728 * Let's pretend the server just sent us a blank line.
2730 snprintf(buf, sizeof(buf), "\r\n");
2731 len = (int)strlen(buf);
2734 * Now, let the normal header parsing algorithm below do its
2735 * job. If it fails, we'll exit instead of continuing.
2742 * If this is an SSL connection or we're in the body
2743 * of the server document, just write it to the client,
2744 * unless we need to buffer the body for later content-filtering
2746 if (server_body || http->ssl)
2751 * If there is no memory left for buffering the content, or the buffer limit
2752 * has been reached, switch to non-filtering mode, i.e. make & write the
2753 * header, flush the iob and buf, and get out of the way.
2755 if (add_to_iob(csp, buf, len))
2760 log_error(LOG_LEVEL_INFO,
2761 "Flushing header and buffers. Stepping back from filtering.");
2763 hdr = list_to_text(csp->headers);
2767 * Memory is too tight to even generate the header.
2768 * Send our static "Out-of-memory" page.
2770 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2771 rsp = cgi_error_memory();
2772 send_crunch_response(csp, rsp);
2775 hdrlen = strlen(hdr);
2777 if (write_socket(csp->cfd, hdr, hdrlen)
2778 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2779 || (write_socket(csp->cfd, buf, (size_t)len)))
2781 log_error(LOG_LEVEL_CONNECT,
2782 "Flush header and buffers to client failed: %E");
2788 * Reset the byte_count to the amount of bytes
2789 * we just flushed. len will be added a few lines below,
2790 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2792 byte_count = (size_t)flushed;
2794 content_filter = NULL;
2800 if (write_socket(csp->cfd, buf, (size_t)len))
2802 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2806 byte_count += (size_t)len;
2811 const char *header_start;
2813 * We're still looking for the end of the server's header.
2814 * Buffer up the data we just read. If that fails, there's
2815 * little we can do but send our static out-of-memory page.
2817 if (add_to_iob(csp, buf, len))
2819 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2820 rsp = cgi_error_memory();
2821 send_crunch_response(csp, rsp);
2825 header_start = csp->iob->cur;
2827 /* Convert iob into something sed() can digest */
2828 if (JB_ERR_PARSE == get_server_headers(csp))
2833 * Well, we tried our MS IIS/5 hack and it didn't work.
2834 * The header is incomplete and there isn't anything
2835 * we can do about it.
2837 log_error(LOG_LEVEL_INFO,
2838 "MS IIS5 hack didn't produce valid headers.");
2844 * Since we have to wait for more from the server before
2845 * we can parse the headers we just continue here.
2851 /* Did we actually get anything? */
2852 if (NULL == csp->headers->first)
2854 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2855 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2856 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2857 free_http_request(http);
2861 assert(csp->headers->first->str);
2863 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2864 strncmpic(csp->headers->first->str, "ICY", 3))
2867 * It doesn't look like a HTTP (or Shoutcast) response:
2868 * tell the client and log the problem.
2870 if (strlen(csp->headers->first->str) > 30)
2872 csp->headers->first->str[30] = '\0';
2874 log_error(LOG_LEVEL_ERROR,
2875 "Invalid server or forwarder response. Starts with: %s",
2876 csp->headers->first->str);
2877 log_error(LOG_LEVEL_CLF,
2878 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2879 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2880 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2881 free_http_request(http);
2886 * We have now received the entire server header,
2887 * filter it and send the result to the client
2889 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2891 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2893 hdr = list_to_text(csp->headers);
2896 /* FIXME Should handle error properly */
2897 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2900 if (crunch_response_triggered(csp, crunchers_light))
2903 * One of the tags created by a server-header
2904 * tagger triggered a crunch. We already
2905 * delivered the crunch response to the client
2906 * and are done here after cleaning up.
2911 /* Buffer and pcrs filter this if appropriate. */
2913 if (!http->ssl) /* We talk plaintext */
2915 content_filter = get_filter_function(csp);
2918 * Only write if we're not buffering for content modification
2920 if (!content_filter)
2923 * Write the server's (modified) header to
2924 * the client (along with anything else that
2925 * may be in the buffer)
2928 if (write_socket(csp->cfd, hdr, strlen(hdr))
2929 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2931 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2934 * The write failed, so don't bother mentioning it
2935 * to the client... it probably can't hear us anyway.
2941 byte_count += (size_t)len;
2946 * XXX: the header lenght should probably
2947 * be calculated by get_server_headers().
2949 int header_length = csp->iob->cur - header_start;
2950 assert(csp->iob->cur > header_start);
2951 byte_count += len - header_length;
2954 /* we're finished with the server's header */
2960 * If this was a MS IIS/5 hack then it means the server
2961 * has already closed the connection. Nothing more to read.
2966 log_error(LOG_LEVEL_INFO,
2967 "Closed server connection detected with MS IIS5 hack enabled.");
2974 * If we reach this point, the server socket is tainted
2975 * (most likely because we didn't read everything the
2976 * server sent us) and reusing it would lead to garbage.
2978 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2980 log_error(LOG_LEVEL_CONNECT, "Unsetting keep-alive flag.");
2981 csp->flags &= ~CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE;
2986 if (csp->content_length == 0)
2989 * If Privoxy didn't recalculate the Content-Lenght,
2990 * byte_count is still correct.
2992 csp->content_length = byte_count;
2995 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2996 csp->ip_addr_str, http->ocmd, csp->content_length);
3000 /*********************************************************************
3004 * Description : This is little more than chat. We only "serve" to
3005 * to close any socket that chat may have opened.
3008 * 1 : csp = Current client state (buffers, headers, etc...)
3012 *********************************************************************/
3014 void serve(struct client_state *csp)
3015 #else /* ifndef AMIGA */
3016 static void serve(struct client_state *csp)
3017 #endif /* def AMIGA */
3020 close_socket(csp->cfd);
3022 if (csp->sfd != JB_INVALID_SOCKET)
3024 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3025 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
3027 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
3031 forget_connection(csp->sfd);
3032 close_socket(csp->sfd);
3035 close_socket(csp->sfd);
3036 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3039 csp->flags &= ~CSP_FLAG_ACTIVE;
3045 /*********************************************************************
3047 * Function : server_thread
3049 * Description : We only exist to call `serve' in a threaded environment.
3052 * 1 : data = Current client state (buffers, headers, etc...)
3054 * Returns : Always 0.
3056 *********************************************************************/
3057 static int32 server_thread(void *data)
3059 serve((struct client_state *) data);
3066 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3067 /*********************************************************************
3071 * Description : Print usage info & exit.
3073 * Parameters : Pointer to argv[0] for identifying ourselves
3077 *********************************************************************/
3078 static void usage(const char *myname)
3080 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3084 #endif /* defined(unix) */
3087 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3088 #endif /* defined(unix) */
3089 "[--version] [configfile]\n"
3090 "Aborting\n", myname);
3095 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3098 #ifdef MUTEX_LOCKS_AVAILABLE
3099 /*********************************************************************
3101 * Function : privoxy_mutex_lock
3103 * Description : Locks a mutex.
3106 * 1 : mutex = The mutex to lock.
3108 * Returns : Void. May exit in case of errors.
3110 *********************************************************************/
3111 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3113 #ifdef FEATURE_PTHREAD
3114 int err = pthread_mutex_lock(mutex);
3117 if (mutex != &log_mutex)
3119 log_error(LOG_LEVEL_FATAL,
3120 "Mutex locking failed: %s.\n", strerror(err));
3125 EnterCriticalSection(mutex);
3126 #endif /* def FEATURE_PTHREAD */
3130 /*********************************************************************
3132 * Function : privoxy_mutex_unlock
3134 * Description : Unlocks a mutex.
3137 * 1 : mutex = The mutex to unlock.
3139 * Returns : Void. May exit in case of errors.
3141 *********************************************************************/
3142 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3144 #ifdef FEATURE_PTHREAD
3145 int err = pthread_mutex_unlock(mutex);
3148 if (mutex != &log_mutex)
3150 log_error(LOG_LEVEL_FATAL,
3151 "Mutex unlocking failed: %s.\n", strerror(err));
3156 LeaveCriticalSection(mutex);
3157 #endif /* def FEATURE_PTHREAD */
3161 /*********************************************************************
3163 * Function : privoxy_mutex_init
3165 * Description : Prepares a mutex.
3168 * 1 : mutex = The mutex to initialize.
3170 * Returns : Void. May exit in case of errors.
3172 *********************************************************************/
3173 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3175 #ifdef FEATURE_PTHREAD
3176 int err = pthread_mutex_init(mutex, 0);
3179 printf("Fatal error. Mutex initialization failed: %s.\n",
3184 InitializeCriticalSection(mutex);
3185 #endif /* def FEATURE_PTHREAD */
3187 #endif /* def MUTEX_LOCKS_AVAILABLE */
3189 /*********************************************************************
3191 * Function : initialize_mutexes
3193 * Description : Prepares mutexes if mutex support is available.
3197 * Returns : Void, exits in case of errors.
3199 *********************************************************************/
3200 static void initialize_mutexes(void)
3202 #ifdef MUTEX_LOCKS_AVAILABLE
3204 * Prepare global mutex semaphores
3206 privoxy_mutex_init(&log_mutex);
3207 privoxy_mutex_init(&log_init_mutex);
3208 privoxy_mutex_init(&connection_reuse_mutex);
3211 * XXX: The assumptions below are a bit naive
3212 * and can cause locks that aren't necessary.
3214 * For example older FreeBSD versions (< 6.x?)
3215 * have no gethostbyname_r, but gethostbyname is
3218 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3219 privoxy_mutex_init(&resolver_mutex);
3220 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3222 * XXX: should we use a single mutex for
3223 * localtime() and gmtime() as well?
3225 #ifndef HAVE_GMTIME_R
3226 privoxy_mutex_init(&gmtime_mutex);
3227 #endif /* ndef HAVE_GMTIME_R */
3229 #ifndef HAVE_LOCALTIME_R
3230 privoxy_mutex_init(&localtime_mutex);
3231 #endif /* ndef HAVE_GMTIME_R */
3234 privoxy_mutex_init(&rand_mutex);
3235 #endif /* ndef HAVE_RANDOM */
3236 #endif /* def MUTEX_LOCKS_AVAILABLE */
3240 /*********************************************************************
3244 * Description : Load the config file and start the listen loop.
3245 * This function is a lot more *sane* with the `load_config'
3246 * and `listen_loop' functions; although it stills does
3247 * a *little* too much for my taste.
3250 * 1 : argc = Number of parameters (including $0).
3251 * 2 : argv = Array of (char *)'s to the parameters.
3253 * Returns : 1 if : can't open config file, unrecognized directive,
3254 * stats requested in multi-thread mode, can't open the
3255 * log file, can't open the jar file, listen port is invalid,
3256 * any load fails, and can't bind port.
3258 * Else main never returns, the process must be signaled
3259 * to terminate execution. Or, on Windows, use the
3260 * "File", "Exit" menu option.
3262 *********************************************************************/
3264 int real_main(int argc, const char *argv[])
3266 int main(int argc, const char *argv[])
3270 unsigned int random_seed;
3272 struct passwd *pw = NULL;
3273 struct group *grp = NULL;
3276 char *pre_chroot_nslookup_to_load_resolver = NULL;
3283 #if !defined(_WIN32)
3291 * Parse the command line arguments
3293 * XXX: simply printing usage information in case of
3294 * invalid arguments isn't particularly user friendly.
3296 while (++argc_pos < argc)
3299 /* Check to see if the service must be installed or uninstalled */
3300 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3302 const char *pName = argv[argc_pos] + 9;
3305 exit( (install_service(pName)) ? 0 : 1 );
3307 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3309 const char *pName = argv[argc_pos] + 11;
3312 exit((uninstall_service(pName)) ? 0 : 1);
3314 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3316 bRunAsService = TRUE;
3317 w32_set_service_cwd();
3318 atexit(w32_service_exit_notify);
3321 #endif /* defined(_WIN32) */
3324 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3326 if (strcmp(argv[argc_pos], "--help") == 0)
3331 else if(strcmp(argv[argc_pos], "--version") == 0)
3333 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3339 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3344 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3346 if (++argc_pos == argc) usage(argv[0]);
3347 pidfile = strdup(argv[argc_pos]);
3350 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3352 if (++argc_pos == argc) usage(argv[argc_pos]);
3354 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3357 if (NULL == (grp = getgrnam(p)))
3359 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3363 if (NULL == (pw = getpwnam(argv[argc_pos])))
3365 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3368 if (p != NULL) *--p = '\0';
3371 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3373 if (++argc_pos == argc) usage(argv[0]);
3374 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3377 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3381 #endif /* defined(unix) */
3383 else if (argc_pos + 1 != argc)
3386 * This is neither the last command line
3387 * option, nor was it recognized before,
3388 * therefore it must be invalid.
3394 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3396 configfile = argv[argc_pos];
3399 } /* -END- while (more arguments) */
3402 if ( *configfile != '/' )
3404 char cwd[BUFFER_SIZE];
3406 size_t abs_file_size;
3408 /* make config-filename absolute here */
3409 if (NULL == getcwd(cwd, sizeof(cwd)))
3411 perror("failed to get current working directory");
3416 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3417 basedir = strdup(cwd);
3419 if (NULL == basedir ||
3420 NULL == (abs_file = malloc(abs_file_size)))
3422 perror("malloc failed");
3425 strlcpy(abs_file, basedir, abs_file_size);
3426 strlcat(abs_file, "/", abs_file_size );
3427 strlcat(abs_file, configfile, abs_file_size);
3428 configfile = abs_file;
3430 #endif /* defined unix */
3434 clients->next = NULL;
3436 /* XXX: factor out initialising after the next stable release. */
3439 #elif defined(_WIN32)
3443 /* Prepare mutexes if supported and necessary. */
3444 initialize_mutexes();
3446 /* Enable logging until further notice. */
3447 init_log_module(Argv[0]);
3449 random_seed = (unsigned int)time(NULL);
3451 srandom(random_seed);
3454 #endif /* ifdef HAVE_RANDOM */
3457 * Unix signal handling
3459 * Catch the abort, interrupt and terminate signals for a graceful exit
3460 * Catch the hangup signal so the errlog can be reopened.
3461 * Ignore the broken pipe signals (FIXME: Why?)
3463 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3466 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3467 const int ignored_signals[] = { SIGPIPE, 0 };
3469 for (idx = 0; catched_signals[idx] != 0; idx++)
3471 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3472 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3474 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3475 #endif /* ifdef sun */
3477 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3481 for (idx = 0; ignored_signals[idx] != 0; idx++)
3483 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3485 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3490 #else /* ifdef _WIN32 */
3491 # ifdef _WIN_CONSOLE
3493 * We *are* in a windows console app.
3494 * Print a verbose messages about FAQ's and such
3496 printf("%s", win32_blurb);
3497 # endif /* def _WIN_CONSOLE */
3498 #endif /* def _WIN32 */
3501 /* Initialize the CGI subsystem */
3502 cgi_init_error_messages();
3505 * If runnig on unix and without the --nodaemon
3506 * option, become a daemon. I.e. fork, detach
3507 * from tty and get process group leadership
3520 if ( pid < 0 ) /* error */
3525 else if ( pid != 0 ) /* parent */
3530 * must check for errors
3531 * child died due to missing files aso
3534 wpid = waitpid( pid, &status, WNOHANG );
3543 /* Should be more portable, but not as well tested */
3548 #else /* ndef __FreeBSD__ */
3550 #endif /* ndef __FreeBSD__ */
3551 fd = open("/dev/tty", O_RDONLY);
3554 /* no error check here */
3555 ioctl( fd, TIOCNOTTY,0 );
3560 * stderr (fd 2) will be closed later on, when the
3561 * log file has been parsed.
3568 } /* -END- if (!no_daemon) */
3571 * As soon as we have written the PID file, we can switch
3572 * to the user and group ID indicated by the --user option
3578 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3580 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3586 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3588 /* Read the time zone file from /etc before doing chroot. */
3590 if (NULL != pre_chroot_nslookup_to_load_resolver
3591 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3593 /* Initialize resolver library. */
3594 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3596 if (chroot(pw->pw_dir) < 0)
3598 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3602 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3605 if (setuid(pw->pw_uid))
3607 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3611 char putenv_dummy[64];
3613 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3614 if (putenv(putenv_dummy) != 0)
3616 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3619 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3620 if (putenv(putenv_dummy) != 0)
3622 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3628 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3631 #endif /* defined unix */
3634 /* This will be FALSE unless the command line specified --service
3638 /* Yup, so now we must attempt to establish a connection
3639 * with the service dispatcher. This will only work if this
3640 * process was launched by the service control manager to
3641 * actually run as a service. If this isn't the case, i've
3642 * known it take around 30 seconds or so for the call to return.
3645 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3646 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3648 /* Service has run, and at this point is now being stopped, so just return */
3653 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3655 /* An error occurred. Usually it's because --service was wrongly specified
3656 * and we were unable to connect to the Service Control Dispatcher because
3657 * it wasn't expecting us and is therefore not listening.
3659 * For now, just continue below to call the listen_loop function.
3662 #endif /* def _WIN32 */
3672 /*********************************************************************
3674 * Function : bind_port_helper
3676 * Description : Bind the listen port. Handles logging, and aborts
3680 * 1 : config = Privoxy configuration. Specifies port
3683 * Returns : Port that was opened.
3685 *********************************************************************/
3686 static jb_socket bind_port_helper(struct configuration_spec * config)
3691 if (config->haddr == NULL)
3693 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3698 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3699 config->hport, config->haddr);
3702 result = bind_port(config->haddr, config->hport, &bfd);
3709 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3710 "There may be another Privoxy or some other "
3711 "proxy running on port %d",
3712 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3713 config->hport, config->hport);
3716 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3717 "The hostname is not resolvable",
3718 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3721 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3722 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3725 /* shouldn't get here */
3726 return JB_INVALID_SOCKET;
3729 config->need_bind = 0;
3736 /* Without this simple workaround we get this compiler warning from _beginthread
3737 * warning C4028: formal parameter 1 different from declaration
3739 void w32_service_listen_loop(void *p)
3743 #endif /* def _WIN32 */
3746 /*********************************************************************
3748 * Function : listen_loop
3750 * Description : bind the listen port and enter a "FOREVER" listening loop.
3756 *********************************************************************/
3757 static void listen_loop(void)
3759 struct client_state *csp = NULL;
3761 struct configuration_spec * config;
3763 config = load_config();
3765 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3767 * XXX: Should be relocated once it no
3768 * longer needs to emit log messages.
3770 initialize_reusable_connections();
3771 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3773 bfd = bind_port_helper(config);
3775 #ifdef FEATURE_GRACEFUL_TERMINATION
3776 while (!g_terminate)
3781 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3782 while (waitpid(-1, NULL, WNOHANG) > 0)
3784 /* zombie children */
3786 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3789 * Free data that was used by died threads
3795 * Re-open the errlog after HUP signal
3797 if (received_hup_signal)
3799 init_error_log(Argv[0], config->logfile);
3800 received_hup_signal = 0;
3804 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3806 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3810 csp->flags |= CSP_FLAG_ACTIVE;
3811 csp->sfd = JB_INVALID_SOCKET;
3813 csp->config = config = load_config();
3815 if ( config->need_bind )
3818 * Since we were listening to the "old port", we will not see
3819 * a "listen" param change until the next IJB request. So, at
3820 * least 1 more request must be made for us to find the new
3821 * setting. I am simply closing the old socket and binding the
3824 * Which-ever is correct, we will serve 1 more page via the
3825 * old settings. This should probably be a "show-proxy-args"
3826 * request. This should not be a so common of an operation
3827 * that this will hurt people's feelings.
3832 bfd = bind_port_helper(config);
3835 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3837 if (!accept_connection(csp, bfd))
3839 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3852 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3855 #ifdef FEATURE_TOGGLE
3856 if (global_toggle_state)
3857 #endif /* def FEATURE_TOGGLE */
3859 csp->flags |= CSP_FLAG_TOGGLED_ON;
3862 if (run_loader(csp))
3864 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3865 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3869 if (block_acl(NULL,csp))
3871 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3872 close_socket(csp->cfd);
3876 #endif /* def FEATURE_ACL */
3878 /* add it to the list of clients */
3879 csp->next = clients->next;
3880 clients->next = csp;
3882 if (config->multi_threaded)
3886 /* this is a switch () statment in the C preprocessor - ugh */
3887 #undef SELECTED_ONE_OPTION
3889 /* Use Pthreads in preference to native code */
3890 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3891 #define SELECTED_ONE_OPTION
3893 pthread_t the_thread;
3894 pthread_attr_t attrs;
3896 pthread_attr_init(&attrs);
3897 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3898 errno = pthread_create(&the_thread, &attrs,
3899 (void * (*)(void *))serve, csp);
3900 child_id = errno ? -1 : 0;
3901 pthread_attr_destroy(&attrs);
3905 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3906 #define SELECTED_ONE_OPTION
3907 child_id = _beginthread(
3908 (void (*)(void *))serve,
3913 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3914 #define SELECTED_ONE_OPTION
3915 child_id = _beginthread(
3916 (void(* _Optlink)(void*))serve,
3922 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3923 #define SELECTED_ONE_OPTION
3925 thread_id tid = spawn_thread
3926 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3928 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3930 child_id = (int) tid;
3939 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3940 #define SELECTED_ONE_OPTION
3941 csp->cfd = ReleaseSocket(csp->cfd, -1);
3944 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3945 NP_Output, Output(),
3946 NP_CloseOutput, FALSE,
3947 NP_Name, (ULONG)"privoxy child",
3951 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3952 NP_Output, Output(),
3953 NP_CloseOutput, FALSE,
3954 NP_Name, (ULONG)"privoxy child",
3955 NP_StackSize, 200*1024,
3961 ((struct Task *)child_id)->tc_UserData = csp;
3962 Signal((struct Task *)child_id, SIGF_SINGLE);
3967 #if !defined(SELECTED_ONE_OPTION)
3970 /* This block is only needed when using fork().
3971 * When using threads, the server thread was
3972 * created and run by the call to _beginthread().
3974 if (child_id == 0) /* child */
3977 #ifdef FEATURE_TOGGLE
3978 int inherited_toggle_state = global_toggle_state;
3979 #endif /* def FEATURE_TOGGLE */
3984 * If we've been toggled or we've blocked the request, tell Mom
3987 #ifdef FEATURE_TOGGLE
3988 if (inherited_toggle_state != global_toggle_state)
3990 rc |= RC_FLAG_TOGGLED;
3992 #endif /* def FEATURE_TOGGLE */
3994 #ifdef FEATURE_STATISTICS
3995 if (csp->flags & CSP_FLAG_REJECTED)
3997 rc |= RC_FLAG_BLOCKED;
3999 #endif /* ndef FEATURE_STATISTICS */
4003 else if (child_id > 0) /* parent */
4005 /* in a fork()'d environment, the parent's
4006 * copy of the client socket and the CSP
4010 #if !defined(_WIN32) && !defined(__CYGWIN__)
4012 wait( &child_status );
4015 * Evaluate child's return code: If the child has
4016 * - been toggled, toggle ourselves
4017 * - blocked its request, bump up the stats counter
4020 #ifdef FEATURE_TOGGLE
4021 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
4023 global_toggle_state = !global_toggle_state;
4025 #endif /* def FEATURE_TOGGLE */
4027 #ifdef FEATURE_STATISTICS
4029 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4033 #endif /* def FEATURE_STATISTICS */
4035 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4036 close_socket(csp->cfd);
4037 csp->flags &= ~CSP_FLAG_ACTIVE;
4041 #undef SELECTED_ONE_OPTION
4042 /* end of cpp switch () */
4044 if (child_id < 0) /* failed */
4046 char buf[BUFFER_SIZE];
4048 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
4050 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
4052 write_socket(csp->cfd, buf, strlen(buf));
4053 close_socket(csp->cfd);
4054 csp->flags &= ~CSP_FLAG_ACTIVE;
4065 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4067 /* Clean up. Aim: free all memory (no leaks) */
4068 #ifdef FEATURE_GRACEFUL_TERMINATION
4070 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4072 unload_current_config_file();
4073 unload_current_actions_file();
4074 unload_current_re_filterfile();
4075 #ifdef FEATURE_TRUST
4076 unload_current_trust_file();
4079 if (config->multi_threaded)
4086 } while ((clients->next != NULL) && (--i > 0));
4090 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4101 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4102 /* Cleanup - remove taskbar icon etc. */
4107 #endif /* FEATURE_GRACEFUL_TERMINATION */