1 const char jcc_rcs[] = "$Id: jcc.c,v 1.193 2008/10/12 15:57:35 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 1.193 2008/10/12 15:57:35 fabiankeil
37 * Fix content length calculation if we read headers
38 * and the start of the body at once. Now that we have
39 * FEATURE_CONNECTION_KEEP_ALIVE, it actually matters.
41 * Revision 1.192 2008/10/11 18:19:14 fabiankeil
42 * Even more chat() cosmetics.
44 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
45 * Reformat some comments in chat().
47 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
48 * In case of chunk-encoded content, stop reading if
49 * the buffer looks like it ends with the last chunk.
51 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
52 * Let server_response_is_complete() deal properly with
53 * content that is neither buffered nor read all at once.
55 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
56 * Flush work-in-progress changes to keep outgoing connections
57 * alive where possible. Incomplete and mostly #ifdef'd out.
59 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
60 * Add mutex lock support for _WIN32.
62 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
63 * Prepare for critical sections on Windows by adding a
64 * layer of indirection before the pthread mutex functions.
66 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
67 * Remove FEATURE_COOKIE_JAR.
69 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
70 * - Silence LLVM/Clang complaint.
71 * - Make received_hup_signal static.
72 * - Hide definitions for basedir, pidfile and received_hup_signal
73 * from __EMX__ as they only seem to be used in case of #ifdef unix.
75 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
76 * Accept Shoutcast responses again. Problem reported
77 * and fix suggested by Stefan in #2062860.
79 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
80 * Fix possible NULL-pointer dereference reported
81 * by din_a4 in #2003937. Pointy hat to me.
83 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
84 * Streamline sed()'s prototype and declare
85 * the header parse and add structures static.
87 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
88 * - Mark csp as immutable for send_crunch_response().
89 * - Fix comment spelling.
91 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
92 * Factor update_server_headers() out of sed(), ditch the
93 * first_run hack and make server_patterns_light static.
95 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
96 * Don't provide get_header() with the whole client state
97 * structure when it only needs access to csp->iob.
99 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
100 * Make the "read the rest of the headers" loop a bit more readable.
102 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
103 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
104 * - Remove useless comment.
106 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
107 * Fix comment grammar.
109 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
110 * Remove the pointless buffer in client_protocol_is_unsupported().
112 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
113 * Least-effort fix for bug #1821930 (reported by Lee):
114 * If the response doesn't look like HTTP,
115 * tell the client and log the problem.
117 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
118 * Don't pass the whole csp structure to flush_socket()
119 * when it only needs a file descriptor and a buffer.
121 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
122 * Remove kill-popups action.
124 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
125 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
127 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
128 * Remove the treat-forbidden-connects-like-blocks action. We now
129 * use the "blocked" page for forbidden CONNECT requests by default.
131 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
132 * Also use shiny new connect_port_is_forbidden() in jcc.c.
134 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
135 * Rename url_actions() to get_url_actions() and let it
136 * use the standard parameter ordering.
138 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
139 * Let forward_url() use the standard parameter ordering
140 * and mark its second parameter immutable.
142 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
143 * Remove the "Listening ... for local connections only" log message.
144 * Whether or not remote connections are able to reach Privoxy is up
145 * to the operating system.
147 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
148 * Prevent the log messages for CONNECT requests to unacceptable
149 * ports from printing the limit-connect argument as [null] if
150 * limit-connect hasn't been explicitly enabled.
152 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
153 * Make sure all console-mode apps get a usage() instance
155 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
156 * Reword NO_SERVER_DATA_RESPONSE to make it harder
157 * to misunderstand what the message is all about.
159 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
160 * Unbreak trustfile which previously didn't work without
161 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
163 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
164 * Plug memory leak. Spotted by Valgrind, triggered by
165 * Privoxy-Regression-Test feeding proxyfuzz.py.
167 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
168 * In the HTTP snipplets, refer to the client as client.
170 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
171 * Emit a log message when activating the MS IIS5 hack.
173 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
174 * Log the "weak randomization factor" warning only
175 * once for mingw32 and provide some more details.
177 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
178 * Initialize log module after initializing mutexes, future
179 * deadlocks in that code should now work cross-platform.
181 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
182 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
183 * Reported by Bert van Leeuwen in BR#1818808.
185 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
186 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
188 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
189 * When in daemon mode, close stderr after the configuration file has been
190 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
192 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
193 * - Fix a crash when parsing invalid requests whose first header
194 * is rejected by get_header(). Regression (re?)introduced
195 * in r1.143 by yours truly.
196 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
197 * client_cookie_adder() to make sure send-vanilla-wafer can be
198 * controlled through tags (and thus regression-tested).
200 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
201 * - Move get_filter_function() from jcc.c to filters.c
202 * so the filter functions can be static.
203 * - Don't bother filtering body-less responses.
205 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
206 * Execute content filters through execute_content_filter().
208 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
209 * Initialize req to NULL to make sure it's defined if the
210 * first read_socket() call fails. Reported by icmp30.
212 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
213 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
214 * extensive comments moved to user manual.
216 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
217 * Don't crash if a broken header filter wiped out the request line.
219 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
220 * Fix byte_count calculation in case of flushes
221 * and don't parse the server headers a second time.
223 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
224 * - If there's a connection problem after we already forwarded
225 * parts of the original content, just hang up. Fixes BR#1776724.
226 * - Fix warnings about unused code on mingw32.
227 * - In case of flushes, calculate the byte count
228 * less incorrectly (I think).
230 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
231 * Add some more prototypes for static functions.
233 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
234 * Comment out request_contains_null_bytes() until it's used again.
236 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
237 * #1763173 from Stefan Huehner: s@const static@static const@
238 * and declare some more functions static.
240 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
241 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
242 * and explain why they were rejected in the first place.
243 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
244 * CONNECT requests. The request line was missing.
245 * - Add two more XXX reminders as we don't have enough already.
247 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
248 * As Hal noticed, checking dispatch_cgi() as the last cruncher
249 * looks like a bug if CGI requests are blocked unintentionally,
250 * so don't do it unless the user enabled the new config option
251 * "allow-cgi-request-crunching".
253 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
254 * - Allow to rewrite the request destination behind the client's back.
255 * - Turn the weird-looking unconditional for loop that
256 * reads the client request into a conditional while loop.
257 * Move the stuff that only runs once out of the loop.
258 * - Move parts of chat(), server_content_type() and the
259 * necessary stuff to fix BR#1750917 into get_filter_function().
261 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
262 * Temporary workaround for BR#1730105.
264 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
265 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
266 * deadlocks and crashes on OpenBSD and possibly other OS with neither
267 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
268 * Thanks to Ralf Horstmann for report and solution.
270 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
271 * Add forward-override{} to change the forwarding settings through
272 * action sections. This is mainly interesting to forward different
273 * clients differently (for example based on User-Agent or request
276 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
277 * - Let usage() mention the --chroot parameter.
278 * - Use read_socket() consistently and always leave
279 * the last buffer byte alone, even in cases where
280 * null termination (currently) doesn't matter.
282 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
283 * - Fix config file loading on Unix if no config file is specified.
284 * Since r1.97 Privoxy would always interpret the last argument as
285 * config file, even if it's a valid command line option.
286 * - Abort in case of unrecognized command line options. Closes #1719696.
287 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
288 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
290 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
291 * - Don't rerun crunchers that only depend on the request URL.
292 * - Don't count redirects and CGI requests as "blocked requests".
294 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
295 * Support crunching based on tags created by server-header taggers.
297 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
298 * Make HTTP snippets static (again). Add a Content-Type for those
299 * with content so the browser doesn't guess it based on the URL.
301 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
302 * Move crunching and request line rebuilding out of chat().
304 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
305 * Introduce tags as alternative way to specify which
306 * actions apply to a request. At the moment tags can be
307 * created based on client and server headers.
309 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
310 * Don't CLF-log CONNECT requests twice.
312 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
313 * Log the source address for ACL-related connection drops.
315 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
316 * New config option: enforce-blocks.
318 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
319 * - Move null byte check into separate function.
320 * - Don't confuse the client with error pages
321 * if a CONNECT request was already confirmed.
323 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
324 * Speed up NULL byte escaping and only log the complete
325 * NULL byte requests with header debugging enabled.
327 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
328 * Answer requests that contain NULL bytes with
329 * a custom response instead of waiting for more
330 * data until the client eventually hangs up.
332 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
333 * - Move delivery and logging of crunched responses
334 * from chat() into send_crunch_response().
335 * - Display the reason for generating http_responses.
336 * - Log the content length for LOG_LEVEL_CLF correctly
337 * (still incorrect for some fixed responses).
338 * - Reword an incorrect comment about
339 * treat-forbidden-connects-like-blocks violating
341 * - Add some log messages.
343 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
344 * Move mutex initialization into separate
345 * function and exit in case of errors.
347 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
348 * - Start to reduce chat()'s line count and move
349 * parts of it into separate functions.
350 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
352 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
353 * - Add Proxy-Agent header to HTTP snippets that are
354 * supposed to reach HTTP clients only.
355 * - Made a few CONNECT log messages more descriptive.
356 * - Catch completely empty server responses (as seen
357 * with Tor's fake ".noconnect" top level domain).
358 * - Use shiny new "forwarding-failed" template for socks errors.
360 * Revision 1.118 2007/01/07 07:43:43 joergs
361 * AmigaOS4 support added.
363 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
364 * Added config option accept-intercepted-requests
365 * and disabled it by default.
367 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
368 * Reverted parts of my last commit
369 * to keep error handling working.
371 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
372 * Fixed gcc43 conversion warnings.
374 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
375 * Fix -pedantic ISO C warning about converting
376 * from function pointer to object pointer.
378 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
379 * Silence compiler warning I introduced with my last commit.
381 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
382 * Mutex protect rand() if POSIX threading
383 * is used, warn the user if that's not possible
384 * and stop using it on _WIN32 where it could
387 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
388 * Don't prevent core dumps by catching SIGABRT.
389 * It's rude and makes debugging unreasonable painful.
391 * Revision 1.110 2006/12/13 14:52:53 etresoft
392 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
394 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
395 * Privoxy is now able to run as intercepting
396 * proxy in combination with any packet filter
397 * that does the port redirection. The destination
398 * is extracted from the "Host:" header which
399 * should be available for nearly all requests.
401 * Moved HTTP snipplets into jcc.c.
402 * Added error message for gopher proxy requests.
404 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
405 * Only unlink the pidfile if it's actually used.
407 * Change order of interception checks to make
408 * it possible to block or redirect requests for
411 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
412 * Make pthread mutex locking more generic. Instead of
413 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
414 * and use mutex locking unless there is an _r function
415 * available. Better safe than sorry.
417 * Fixes "./configure --disable-pthread" and should result
418 * in less threading-related problems on pthread-using platforms,
419 * but it still doesn't fix BR#1122404.
421 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
422 * Move pthread.h inclusion from jcc.c to jcc.h.
423 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
425 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
426 * Don't exit after receiving the second SIGHUP on Solaris.
428 * Fixes BR 1052235, but the same problem may exist on other
429 * systems. Once 3.0.6 is out we should use sigset()
430 * where available and see if it breaks anything.
432 * Revision 1.104 2006/09/23 13:26:38 roro
433 * Replace TABs by spaces in source code.
435 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
436 * Fix +redirect{}. Didn't work with -fast-redirects.
438 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
439 * Respond with 400 and a short text message
440 * if the client tries to use Privoxy as FTP proxy.
442 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
443 * Make number of retries in case of forwarded-connect problems
444 * a config file option (forwarded-connect-retries) and use 0 as
447 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
448 * Set random(3) seed.
450 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
451 * Follow the OpenBSD port's lead and protect the resolve
452 * functions on OpenBSD as well.
454 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
455 * --user fix. Only use the user as group if no group is specified.
456 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
458 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
459 * Windows service (re-)integration
463 * --install[:service_name]
464 * --uninstall[:service_name]
467 * They work as follows:
468 * --install will create a service for you and then terminate.
469 * By default the service name will be "privoxy" (without the quotes).
470 * However you can run multiple services if you wish, just by adding
471 * a colon and then a name (no spaces).
473 * --uninstall follows the exact same rules a --install.
475 * --service is used when the program is executed by the service
476 * control manager, and in normal circumstances would never be
477 * used as a command line argument.
479 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
480 * Windows service integration
482 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
483 * Incorporate Fabian Keil's patch work:
484 http://www.fabiankeil.de/sourcecode/privoxy/
486 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
487 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
488 * with what was really the latest development (the v_3_0_branch branch)
490 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
491 * Thanks to Jindrich Makovicka for a race condition fix for the log
492 * file. The race condition remains for non-pthread implementations.
493 * Reference patch #1175720.
495 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
496 * Add the ability to check jpeg images for invalid
497 * lengths of comment blocks. Defensive strategy
498 * against the exploit:
499 * Microsoft Security Bulletin MS04-028
500 * Buffer Overrun in JPEG Processing (GDI+) Could
501 * Allow Code Execution (833987)
502 * Enabled with +inspect-jpegs in actions files.
504 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
505 * - Fixed usage info for non-unix platforms
506 * - Fixed small cmdline parsing bug
508 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
509 * Diagnostics: Now preserve the returncode of pthread_create
510 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
512 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
513 * No longer ignore SIGCHLD. Fixes bug #769381
515 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
516 * Close jarfile on graceful exit, remove stray line
518 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
519 * Cosmetics: Killed a warning, a typo and an allocation left at exit
521 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
522 * No longer rely on non-POSIX.1 extensions of getcwd().
525 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
526 * Replaced setenv() by posix-compliant putenv()
527 * Thanks to Neil McCalden (nmcc AT users.sf.net).
529 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
530 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
532 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
533 * Clean-up and extension of improvements for forked mode:
534 * - Child's return code now consists of flags RC_FLAG_*
535 * - Reporting toggle to parent now properly #ifdef'ed
536 * - Children now report blocking to parent. This enables
537 * statistics in forked mode
539 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
540 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
541 * Children now report having been toggled through _exit(17), parents
542 * watch for that code and toggle themselves if found.
544 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
545 * Wrapping all *_r functions (the non-_r versions of them) with
546 * mutex semaphores for OSX. Hopefully this will take care of all
547 * of those pesky crash reports.
549 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
550 * Fixed two mostly harmless mem leaks
552 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
553 * Fix: Head of global clients list now initialized to NULL
555 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
556 * Added basic support for OPTIONS and TRACE HTTP methods:
557 * - New interceptor direct_response() added in chat().
558 * - sed() moved to earlier in the process, so that the
559 * Host: header is evaluated before actions and forwarding
562 * Revision 1.92 2002/05/08 16:00:46 oes
563 * Chat's buffer handling:
564 * - Fixed bug with unchecked out-of-mem conditions
565 * while reading client request & server headers
566 * - No longer predict if the buffer limit will be exceeded
567 * in the next read -- check add_to_iob's new
568 * return code. If buffer couldn't be extended
569 * (policy or out-of-mem) while
570 * - reading from client: abort
571 * - reading server headers: send error page
572 * - buffering server body for filter: flush,
573 * and if that fails: send error page
575 * Revision 1.91 2002/04/08 20:35:58 swa
578 * Revision 1.90 2002/04/02 14:57:28 oes
579 * Made sending wafers independent of FEATURE_COOKIE_JAR
581 * Revision 1.89 2002/03/31 17:18:59 jongfoster
582 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
584 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
585 * More compiler warning message maintenance
587 * Revision 1.87 2002/03/26 22:29:54 swa
588 * we have a new homepage!
590 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
591 * Workaround for closing the jarfile before load_config() comes around again
593 * Revision 1.85 2002/03/24 15:23:33 jongfoster
596 * Revision 1.84 2002/03/24 13:25:43 swa
597 * name change related issues
599 * Revision 1.83 2002/03/16 23:54:06 jongfoster
600 * Adding graceful termination feature, to help look for memory leaks.
601 * If you enable this (which, by design, has to be done by hand
602 * editing config.h) and then go to http://i.j.b/die, then the program
603 * will exit cleanly after the *next* request. It should free all the
604 * memory that was used.
606 * Revision 1.82 2002/03/13 00:27:05 jongfoster
609 * Revision 1.81 2002/03/12 01:42:50 oes
610 * Introduced modular filters
612 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
613 * OS/2 port maintenance:
614 * - Fixed EMX build - it had decayed a little
615 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
616 * substituted a memset for now.
618 * Revision 1.79 2002/03/09 20:03:52 jongfoster
619 * - Making various functions return int rather than size_t.
620 * (Undoing a recent change). Since size_t is unsigned on
621 * Windows, functions like read_socket that return -1 on
622 * error cannot return a size_t.
624 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
625 * crashes, and also frequently caused JB to jump to 100%
626 * CPU and stay there. (Because it thought it had just
627 * read ((unsigned)-1) == 4Gb of data...)
629 * - The signature of write_socket has changed, it now simply
630 * returns success=0/failure=nonzero.
632 * - Trying to get rid of a few warnings --with-debug on
633 * Windows, I've introduced a new type "jb_socket". This is
634 * used for the socket file descriptors. On Windows, this
635 * is SOCKET (a typedef for unsigned). Everywhere else, it's
636 * an int. The error value can't be -1 any more, so it's
637 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
638 * Windows it maps to the #define INVALID_SOCKET.)
640 * - The signature of bind_port has changed.
642 * Revision 1.78 2002/03/08 21:35:04 oes
643 * Added optional group supplement to --user option. Will now use default group of user if no group given
645 * Revision 1.77 2002/03/07 03:52:06 oes
646 * - Fixed compiler warnings etc
647 * - Improved handling of failed DNS lookups
649 * Revision 1.76 2002/03/06 22:54:35 jongfoster
650 * Automated function-comment nitpicking.
652 * Revision 1.75 2002/03/06 10:02:19 oes
653 * Fixed stupid bug when --user was not given
655 * Revision 1.74 2002/03/06 00:49:31 jongfoster
656 * Fixing warning on Windows
657 * Making #ifdefs that refer to the same variable consistently
658 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
660 * Revision 1.73 2002/03/05 23:57:30 hal9
661 * Stray character 's' on line 1618 was breaking build.
663 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
664 * - Re-enable OS/2 building after new parms were added
665 * - Fix false out of memory report when resolving CGI templates when no IP
666 * address is available of failed attempt (a la no such domain)
668 * Revision 1.71 2002/03/05 18:13:56 oes
669 * Added --user option
671 * Revision 1.70 2002/03/05 04:52:42 oes
672 * Deleted non-errlog debugging code
674 * Revision 1.69 2002/03/04 23:50:00 jongfoster
675 * Splitting off bind_port() call into bind_port_helper(), with
678 * Revision 1.68 2002/03/04 20:17:32 oes
681 * Revision 1.67 2002/03/04 18:18:57 oes
682 * - Removed _DEBUG mode
683 * - Cleand up cmdline parsing
684 * - Introduced --no-daemon, --pidfile options
685 * - Cleaned up signal handling:
686 * - Terminate cleanly on INT, TERM and ABRT
687 * - Schedule logfile for re-opening on HUP
688 * - Ignore CHLD and PIPE
689 * - Leave the rest with their default handlers
690 * - Uniform handler registration
691 * - Added usage() function
692 * - Played styleguide police
694 * Revision 1.66 2002/03/03 15:06:55 oes
695 * Re-enabled automatic config reloading
697 * Revision 1.65 2002/03/03 14:49:11 oes
698 * Fixed CLF logging: Now uses client's original HTTP request
700 * Revision 1.64 2002/03/03 09:18:03 joergs
701 * Made jumbjuster work on AmigaOS again.
703 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
704 * Clean up a little CRLF unpleasantness that suddenly appeared
706 * Revision 1.62 2002/02/20 23:17:23 jongfoster
707 * Detecting some out-of memory conditions and exiting with a log message.
709 * Revision 1.61 2002/01/17 21:01:52 jongfoster
710 * Moving all our URL and URL pattern parsing code to urlmatch.c.
712 * Revision 1.60 2001/12/30 14:07:32 steudten
713 * - Add signal handling (unix)
714 * - Add SIGHUP handler (unix)
715 * - Add creation of pidfile (unix)
716 * - Add action 'top' in rc file (RH)
717 * - Add entry 'SIGNALS' to manpage
718 * - Add exit message to logfile (unix)
720 * Revision 1.59 2001/12/13 14:07:18 oes
721 * Fixed Bug: 503 error page now sent OK
723 * Revision 1.58 2001/11/30 23:37:24 jongfoster
724 * Renaming the Win32 config file to config.txt - this is almost the
725 * same as the corresponding UNIX name "config"
727 * Revision 1.57 2001/11/16 00:47:43 jongfoster
728 * Changing the tty-disconnection code to use setsid().
730 * Revision 1.56 2001/11/13 20:20:54 jongfoster
731 * Tabs->spaces, fixing a bug with missing {} around an if()
733 * Revision 1.55 2001/11/13 20:14:53 jongfoster
734 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
736 * Revision 1.54 2001/11/07 00:03:14 steudten
737 * Give reliable return value if an error
738 * occurs not just 0 with new daemon mode.
740 * Revision 1.53 2001/11/05 21:41:43 steudten
741 * Add changes to be a real daemon just for unix os.
742 * (change cwd to /, detach from controlling tty, set
743 * process group and session leader to the own process.
745 * Add some fatal-error log message for failed malloc().
746 * Add '-d' if compiled with 'configure --with-debug' to
747 * enable debug output.
749 * Revision 1.52 2001/10/26 20:11:20 jongfoster
750 * Fixing type mismatch
752 * Revision 1.51 2001/10/26 17:38:28 oes
755 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
756 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
757 * threads to call select() simultaneously. So, it's time to do a real, live,
758 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
759 * (native). Both versions will work, but using __OS2__ offers multi-threading.
761 * Revision 1.49 2001/10/23 21:41:35 jongfoster
762 * Added call to initialize the (statically-allocated of course)
763 * "out of memory" CGI response.
765 * Revision 1.48 2001/10/10 19:56:46 jongfoster
766 * Moving some code that wasn't cookie-related out of an #ifdef
769 * Revision 1.47 2001/10/10 16:44:36 oes
770 * Added CONNECT destination port limitation check
772 * Revision 1.46 2001/10/08 15:17:41 oes
773 * Re-enabled SSL forwarding
775 * Revision 1.45 2001/10/07 15:42:11 oes
776 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
778 * Moved downgrading of the HTTP version from parse_http_request to
779 * chat(), since we can't decide if it is necessary before we have
780 * determined the actions for the URL. The HTTP command is now
781 * *always* re-built so the repairs need no longer be special-cased.
783 * filter_popups now gets a csp pointer so it can raise the new
784 * CSP_FLAG_MODIFIED flag.
788 * Added configurable size limit for the IOB. If the IOB grows so
789 * large that the next read would exceed the limit, the header
790 * is generated, and the header & unfiltered buffer are flushed
791 * to the client. Chat then continues in non-buffering,
792 * non-filtering body mode.
794 * Revision 1.44 2001/10/02 18:13:57 oes
797 * Revision 1.43 2001/10/02 15:32:13 oes
798 * Moved generation of hdr
800 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
801 * Cleaning up 2 compiler warnings on OS/2.
803 * Revision 1.41 2001/09/16 17:05:14 jongfoster
804 * Removing unused #include showarg.h
806 * Revision 1.40 2001/09/16 15:41:45 jongfoster
807 * Fixing signed/unsigned comparison warning.
809 * Revision 1.39 2001/09/16 13:21:27 jongfoster
810 * Changes to use new list functions.
812 * Revision 1.38 2001/09/16 13:01:46 jongfoster
813 * Removing redundant function call that zeroed zalloc()'d memory.
815 * Revision 1.37 2001/09/10 11:12:24 oes
816 * Deleted unused variable
818 * Revision 1.36 2001/09/10 10:56:15 oes
819 * Silenced compiler warnings
821 * Revision 1.35 2001/07/31 14:44:22 oes
822 * Deleted unused size parameter from filter_popups()
824 * Revision 1.34 2001/07/30 22:08:36 jongfoster
825 * Tidying up #defines:
826 * - All feature #defines are now of the form FEATURE_xxx
827 * - Permanently turned off WIN_GUI_EDIT
828 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
830 * Revision 1.33 2001/07/29 19:32:00 jongfoster
831 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
833 * Revision 1.32 2001/07/29 18:47:05 jongfoster
834 * Adding missing #include "loadcfg.h"
836 * Revision 1.31 2001/07/29 12:17:48 oes
837 * Applied pthread fix by Paul Lieverse
839 * Revision 1.30 2001/07/25 22:57:13 jongfoster
840 * __BEOS__ no longer overrides FEATURE_PTHREAD.
841 * This is because FEATURE_PTHREAD will soon be widely used, so I
842 * want to keep it simple.
844 * Revision 1.29 2001/07/24 12:47:06 oes
845 * Applied BeOS support update by Eugenia
847 * Revision 1.28 2001/07/23 13:26:12 oes
848 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
850 * Revision 1.27 2001/07/19 19:09:47 haroon
851 * - Added code to take care of the situation where while processing the first
852 * server response (which includes the server header), after finding the end
853 * of the headers we were not looking past the end of the headers for
854 * content modification. I enabled it for filter_popups.
855 * Someone else should look to see if other similar operations should be
856 * done to the discarded portion of the buffer.
858 * Note 2001/07/20: No, the other content modification mechanisms will process
859 * the whole iob later anyway. --oes
861 * Revision 1.26 2001/07/18 12:31:36 oes
864 * Revision 1.25 2001/07/15 19:43:49 jongfoster
865 * Supports POSIX threads.
866 * Also removed some unused #includes.
868 * Revision 1.24 2001/07/13 14:00:40 oes
869 * - Generic content modification scheme:
870 * Each feature has its own applicability flag that is set
871 * from csp->action->flags.
872 * Replaced the "filtering" int flag , by a function pointer
873 * "content_filter" to the function that will do the content
874 * modification. If it is != NULL, the document will be buffered
875 * and processed through *content_filter, which must set
876 * csp->content_length and return a modified copy of the body
877 * or return NULL (on failiure).
878 * - Changed csp->is_text to the more generic bitmap csp->content_type
879 * which can currently take the valued CT_TEXT or CT_GIF
881 * - Removed all #ifdef PCRS
883 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
884 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
885 * the `block_acl' call. This prevents a compilation error when the user
886 * does not wish to use the "ACL" feature.
888 * Revision 1.22 2001/06/29 21:45:41 oes
889 * Indentation, CRLF->LF, Tab-> Space
891 * Revision 1.21 2001/06/29 13:29:36 oes
892 * - Cleaned up, improved comments
893 * - Unified all possible interceptors (CGI,
894 * block, trust, fast_redirect) in one
895 * place, with one (CGI) answer generation
896 * mechansim. Much clearer now.
897 * - Removed the GIF image generation, which
898 * is now done in filters.c:block_url()
899 * - Made error conditions like domain lookup
900 * failiure or (various) problems while talking
901 * to the server use cgi.c:error_response()
902 * instead of generating HTML/HTTP in chat() (yuck!)
903 * - Removed logentry from cancelled commit
905 * Revision 1.20 2001/06/09 10:55:28 jongfoster
906 * Changing BUFSIZ ==> BUFFER_SIZE
908 * Revision 1.19 2001/06/07 23:12:52 jongfoster
909 * Replacing function pointer in struct gateway with a directly
910 * called function forwarded_connect().
911 * Replacing struct gateway with struct forward_spec
913 * Revision 1.18 2001/06/03 19:12:16 oes
914 * introduced new cgi handling
916 * Revision 1.17 2001/06/01 20:07:23 jongfoster
917 * Now uses action +image-blocker{} rather than config->tinygif
919 * Revision 1.16 2001/06/01 18:49:17 jongfoster
920 * Replaced "list_share" with "list" - the tiny memory gain was not
921 * worth the extra complexity.
923 * Revision 1.15 2001/05/31 21:24:47 jongfoster
924 * Changed "permission" to "action" throughout.
925 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
926 * Moved vanilla wafer check into chat(), since we must now
927 * decide whether or not to add it based on the URL.
929 * Revision 1.14 2001/05/29 20:14:01 joergs
930 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
933 * Revision 1.13 2001/05/29 09:50:24 jongfoster
934 * Unified blocklist/imagelist/permissionslist.
935 * File format is still under discussion, but the internal changes
938 * Also modified interceptor behaviour:
939 * - We now intercept all URLs beginning with one of the following
940 * prefixes (and *only* these prefixes):
942 * * http://ijbswa.sf.net/config/
943 * * http://ijbswa.sourceforge.net/config/
944 * - New interceptors "home page" - go to http://i.j.b/ to see it.
945 * - Internal changes so that intercepted and fast redirect pages
946 * are not replaced with an image.
947 * - Interceptors now have the option to send a binary page direct
948 * to the client. (i.e. ijb-send-banner uses this)
949 * - Implemented show-url-info interceptor. (Which is why I needed
950 * the above interceptors changes - a typical URL is
951 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
952 * The previous mechanism would not have intercepted that, and
953 * if it had been intercepted then it then it would have replaced
956 * Revision 1.12 2001/05/27 22:17:04 oes
958 * - re_process_buffer no longer writes the modified buffer
959 * to the client, which was very ugly. It now returns the
960 * buffer, which it is then written by chat.
962 * - content_length now adjusts the Content-Length: header
963 * for modified documents rather than crunch()ing it.
964 * (Length info in csp->content_length, which is 0 for
965 * unmodified documents)
967 * - For this to work, sed() is called twice when filtering.
969 * Revision 1.11 2001/05/26 17:27:53 jongfoster
970 * Added support for CLF and fixed LOG_LEVEL_LOG.
971 * Also did CRLF->LF fix of my previous patch.
973 * Revision 1.10 2001/05/26 15:26:15 jongfoster
974 * ACL feature now provides more security by immediately dropping
975 * connections from untrusted hosts.
977 * Revision 1.9 2001/05/26 00:28:36 jongfoster
978 * Automatic reloading of config file.
979 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
980 * Most of the global variables have been moved to a new
981 * struct configuration_spec, accessed through csp->config->globalname
982 * Most of the globals remaining are used by the Win32 GUI.
984 * Revision 1.8 2001/05/25 22:43:18 jongfoster
985 * Fixing minor memory leak and buffer overflow.
987 * Revision 1.7 2001/05/25 22:34:30 jongfoster
990 * Revision 1.6 2001/05/23 00:13:58 joergs
991 * AmigaOS support fixed.
993 * Revision 1.5 2001/05/22 18:46:04 oes
995 * - Enabled filtering banners by size rather than URL
996 * by adding patterns that replace all standard banner
997 * sizes with the "Junkbuster" gif to the re_filterfile
999 * - Enabled filtering WebBugs by providing a pattern
1000 * which kills all 1x1 images
1002 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
1003 * which is selected by the (nonstandard and therefore
1004 * capital) letter 'U' in the option string.
1005 * It causes the quantifiers to be ungreedy by default.
1006 * Appending a ? turns back to greedy (!).
1008 * - Added a new interceptor ijb-send-banner, which
1009 * sends back the "Junkbuster" gif. Without imagelist or
1010 * MSIE detection support, or if tinygif = 1, or the
1011 * URL isn't recognized as an imageurl, a lame HTML
1012 * explanation is sent instead.
1014 * - Added new feature, which permits blocking remote
1015 * script redirects and firing back a local redirect
1017 * The feature is conditionally compiled, i.e. it
1018 * can be disabled with --disable-fast-redirects,
1019 * plus it must be activated by a "fast-redirects"
1020 * line in the config file, has its own log level
1021 * and of course wants to be displayed by show-proxy-args
1022 * Note: Boy, all the #ifdefs in 1001 locations and
1023 * all the fumbling with configure.in and acconfig.h
1024 * were *way* more work than the feature itself :-(
1026 * - Because a generic redirect template was needed for
1027 * this, tinygif = 3 now uses the same.
1029 * - Moved GIFs, and other static HTTP response templates
1032 * - Some minor fixes
1034 * - Removed some >400 CRs again (Jon, you really worked
1037 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1038 * Made failure to bind() a fatal error.
1040 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1041 * Version 2.9.4 checkin.
1042 * - Merged popupfile and cookiefile, and added control over PCRS
1043 * filtering, in new "permissionsfile".
1044 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1045 * file error you now get a message box (in the Win32 GUI) rather
1046 * than the program exiting with no explanation.
1047 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1049 * - Removed tabs from "config"
1050 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1051 * - Bumped up version number.
1053 * Revision 1.2 2001/05/17 22:34:44 oes
1054 * - Added hint on GIF char array generation to jcc.c
1055 * - Cleaned CRLF's from the sources and related files
1056 * - Repaired logging for REF and FRC
1058 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1059 * Initial import of version 2.9.3 source tree
1062 *********************************************************************/
1068 #include <sys/types.h>
1077 # ifndef FEATURE_PTHREAD
1081 # include <windows.h>
1082 # include <process.h>
1083 # endif /* ndef FEATURE_PTHREAD */
1086 # ifndef _WIN_CONSOLE
1087 # include "w32log.h"
1088 # endif /* ndef _WIN_CONSOLE */
1089 # include "w32svrapi.h"
1091 #else /* ifndef _WIN32 */
1093 # if !defined (__OS2__)
1094 # include <unistd.h>
1095 # include <sys/wait.h>
1096 # endif /* ndef __OS2__ */
1097 # include <sys/time.h>
1098 # include <sys/stat.h>
1099 # include <sys/ioctl.h>
1102 #include <sys/termios.h>
1110 # include <signal.h>
1113 # include <socket.h> /* BeOS has select() for sockets only. */
1114 # include <OS.h> /* declarations for threads and stuff. */
1117 # if defined(__EMX__) || defined(__OS2__)
1118 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1123 #define bzero(B,N) memset(B,0x00,n)
1127 # include <select.h>
1132 #include "project.h"
1135 #include "filters.h"
1136 #include "loaders.h"
1137 #include "parsers.h"
1138 #include "miscutil.h"
1140 #include "jbsockets.h"
1141 #include "gateway.h"
1142 #include "actions.h"
1144 #include "loadcfg.h"
1145 #include "urlmatch.h"
1147 const char jcc_h_rcs[] = JCC_H_VERSION;
1148 const char project_h_rcs[] = PROJECT_H_VERSION;
1151 struct client_state clients[1];
1152 struct file_list files[1];
1154 #ifdef FEATURE_STATISTICS
1155 int urls_read = 0; /* total nr of urls read inc rejected */
1156 int urls_rejected = 0; /* total nr of urls rejected */
1157 #endif /* def FEATURE_STATISTICS */
1159 #ifdef FEATURE_GRACEFUL_TERMINATION
1160 int g_terminate = 0;
1163 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1164 static void sig_handler(int the_signal);
1166 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1167 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1168 static jb_err get_server_headers(struct client_state *csp);
1169 static const char *crunch_reason(const struct http_response *rsp);
1170 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1172 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1174 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1175 static jb_err change_request_destination(struct client_state *csp);
1176 static void chat(struct client_state *csp);
1177 static void serve(struct client_state *csp);
1178 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1179 static void usage(const char *myname);
1181 static void initialize_mutexes(void);
1182 static jb_socket bind_port_helper(struct configuration_spec *config);
1183 static void listen_loop(void);
1186 void serve(struct client_state *csp);
1187 #else /* ifndef AMIGA */
1188 static void serve(struct client_state *csp);
1189 #endif /* def AMIGA */
1192 static int32 server_thread(void *data);
1193 #endif /* def __BEOS__ */
1196 #define sleep(N) Sleep(((N) * 1000))
1200 #define sleep(N) DosSleep(((N) * 100))
1203 #ifdef MUTEX_LOCKS_AVAILABLE
1205 * XXX: Does the locking stuff really belong in this file?
1207 privoxy_mutex_t log_mutex;
1208 privoxy_mutex_t log_init_mutex;
1209 privoxy_mutex_t connection_reuse_mutex;
1211 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1212 privoxy_mutex_t resolver_mutex;
1213 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1215 #ifndef HAVE_GMTIME_R
1216 privoxy_mutex_t gmtime_mutex;
1217 #endif /* ndef HAVE_GMTIME_R */
1219 #ifndef HAVE_LOCALTIME_R
1220 privoxy_mutex_t localtime_mutex;
1221 #endif /* ndef HAVE_GMTIME_R */
1224 privoxy_mutex_t rand_mutex;
1225 #endif /* ndef HAVE_RANDOM */
1227 #endif /* def MUTEX_LOCKS_AVAILABLE */
1230 const char *basedir = NULL;
1231 const char *pidfile = NULL;
1232 static int received_hup_signal = 0;
1233 #endif /* defined unix */
1235 /* HTTP snipplets. */
1236 static const char CSUCCEED[] =
1237 "HTTP/1.0 200 Connection established\r\n"
1238 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1240 static const char CHEADER[] =
1241 "HTTP/1.0 400 Invalid header received from client\r\n"
1242 "Proxy-Agent: Privoxy " VERSION "\r\n"
1243 "Content-Type: text/plain\r\n"
1244 "Connection: close\r\n\r\n"
1245 "Invalid header received from client.\r\n";
1247 static const char FTP_RESPONSE[] =
1248 "HTTP/1.0 400 Invalid request received from client\r\n"
1249 "Content-Type: text/plain\r\n"
1250 "Connection: close\r\n\r\n"
1251 "Invalid request. Privoxy doesn't support FTP.\r\n";
1253 static const char GOPHER_RESPONSE[] =
1254 "HTTP/1.0 400 Invalid request received from client\r\n"
1255 "Content-Type: text/plain\r\n"
1256 "Connection: close\r\n\r\n"
1257 "Invalid request. Privoxy doesn't support gopher.\r\n";
1259 /* XXX: should be a template */
1260 static const char MISSING_DESTINATION_RESPONSE[] =
1261 "HTTP/1.0 400 Bad request received from client\r\n"
1262 "Proxy-Agent: Privoxy " VERSION "\r\n"
1263 "Content-Type: text/plain\r\n"
1264 "Connection: close\r\n\r\n"
1265 "Bad request. Privoxy was unable to extract the destination.\r\n";
1267 /* XXX: should be a template */
1268 static const char NO_SERVER_DATA_RESPONSE[] =
1269 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1270 "Proxy-Agent: Privoxy " VERSION "\r\n"
1271 "Content-Type: text/plain\r\n"
1272 "Connection: close\r\n\r\n"
1273 "Empty server or forwarder response.\r\n"
1274 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1276 /* XXX: should be a template */
1277 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1278 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1279 "Proxy-Agent: Privoxy " VERSION "\r\n"
1280 "Content-Type: text/plain\r\n"
1281 "Connection: close\r\n\r\n"
1282 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1285 /* XXX: should be a template */
1286 static const char NULL_BYTE_RESPONSE[] =
1287 "HTTP/1.0 400 Bad request received from client\r\n"
1288 "Proxy-Agent: Privoxy " VERSION "\r\n"
1289 "Content-Type: text/plain\r\n"
1290 "Connection: close\r\n\r\n"
1291 "Bad request. Null byte(s) before end of request.\r\n";
1294 /* XXX: should be a template */
1295 static const char MESSED_UP_REQUEST_RESPONSE[] =
1296 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1297 "Proxy-Agent: Privoxy " VERSION "\r\n"
1298 "Content-Type: text/plain\r\n"
1299 "Connection: close\r\n\r\n"
1300 "Bad request. Messed up with header filters.\r\n";
1302 /* A function to crunch a response */
1303 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1305 /* Crunch function flags */
1306 #define CF_NO_FLAGS 0
1307 /* Cruncher applies to forced requests as well */
1308 #define CF_IGNORE_FORCE 1
1309 /* Crunched requests are counted for the block statistics */
1310 #define CF_COUNT_AS_REJECT 2
1312 /* A crunch function and its flags */
1315 const crunch_func_ptr cruncher;
1319 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1321 /* Complete list of cruncher functions */
1322 static const struct cruncher crunchers_all[] = {
1323 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1324 { block_url, CF_COUNT_AS_REJECT },
1325 #ifdef FEATURE_TRUST
1326 { trust_url, CF_COUNT_AS_REJECT },
1327 #endif /* def FEATURE_TRUST */
1328 { redirect_url, CF_NO_FLAGS },
1329 { dispatch_cgi, CF_IGNORE_FORCE},
1333 /* Light version, used after tags are applied */
1334 static const struct cruncher crunchers_light[] = {
1335 { block_url, CF_COUNT_AS_REJECT },
1336 { redirect_url, CF_NO_FLAGS },
1342 * XXX: Don't we really mean
1348 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1349 /*********************************************************************
1351 * Function : sig_handler
1353 * Description : Signal handler for different signals.
1354 * Exit gracefully on TERM and INT
1355 * or set a flag that will cause the errlog
1356 * to be reopened by the main thread on HUP.
1359 * 1 : the_signal = the signal cause this function to call
1363 *********************************************************************/
1364 static void sig_handler(int the_signal)
1370 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1382 received_hup_signal = 1;
1388 * We shouldn't be here, unless we catch signals
1389 * in main() that we can't handle here!
1391 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1399 /*********************************************************************
1401 * Function : client_protocol_is_unsupported
1403 * Description : Checks if the client used a known unsupported
1404 * protocol and deals with it by sending an error
1408 * 1 : csp = Current client state (buffers, headers, etc...)
1409 * 2 : req = the first request line send by the client
1411 * Returns : TRUE if an error response has been generated, or
1412 * FALSE if the request doesn't look invalid.
1414 *********************************************************************/
1415 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1418 * If it's a FTP or gopher request, we don't support it.
1420 * These checks are better than nothing, but they might
1421 * not work in all configurations and some clients might
1422 * have problems digesting the answer.
1424 * They should, however, never cause more problems than
1425 * Privoxy's old behaviour (returning the misleading HTML
1428 * "Could not resolve http://(ftp|gopher)://example.org").
1430 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1432 const char *response = NULL;
1433 const char *protocol = NULL;
1435 if (!strncmpic(req, "GET ftp://", 10))
1437 response = FTP_RESPONSE;
1442 response = GOPHER_RESPONSE;
1443 protocol = "GOPHER";
1445 log_error(LOG_LEVEL_ERROR,
1446 "%s tried to use Privoxy as %s proxy: %s",
1447 csp->ip_addr_str, protocol, req);
1448 log_error(LOG_LEVEL_CLF,
1449 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1451 write_socket(csp->cfd, response, strlen(response));
1460 /*********************************************************************
1462 * Function : get_request_destination_elsewhere
1464 * Description : If the client's request was redirected into
1465 * Privoxy without the client's knowledge,
1466 * the request line lacks the destination host.
1468 * This function tries to get it elsewhere,
1469 * provided accept-intercepted-requests is enabled.
1471 * "Elsewhere" currently only means "Host: header",
1472 * but in the future we may ask the redirecting
1473 * packet filter to look the destination up.
1475 * If the destination stays unknown, an error
1476 * response is send to the client and headers
1477 * are freed so that chat() can return directly.
1480 * 1 : csp = Current client state (buffers, headers, etc...)
1481 * 2 : headers = a header list
1483 * Returns : JB_ERR_OK if the destination is now known, or
1484 * JB_ERR_PARSE if it isn't.
1486 *********************************************************************/
1487 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1491 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1493 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1494 " Privoxy isn't configured to accept intercepted requests.",
1495 csp->ip_addr_str, csp->http->cmd);
1496 /* XXX: Use correct size */
1497 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1498 csp->ip_addr_str, csp->http->cmd);
1500 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1501 destroy_list(headers);
1503 return JB_ERR_PARSE;
1505 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1507 /* Split the domain we just got for pattern matching */
1508 init_domain_components(csp->http);
1514 /* We can't work without destination. Go spread the news.*/
1516 req = list_to_text(headers);
1518 /* XXX: Use correct size */
1519 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1520 csp->ip_addr_str, csp->http->cmd);
1521 log_error(LOG_LEVEL_ERROR,
1522 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1523 csp->ip_addr_str, csp->http->cmd, req);
1526 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1527 destroy_list(headers);
1529 return JB_ERR_PARSE;
1532 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1533 * to get the destination IP address, use it as host directly
1534 * or do a reverse DNS lookup first.
1539 /*********************************************************************
1541 * Function : get_server_headers
1543 * Description : Parses server headers in iob and fills them
1544 * into csp->headers so that they can later be
1548 * 1 : csp = Current client state (buffers, headers, etc...)
1550 * Returns : JB_ERR_OK if everything went fine, or
1551 * JB_ERR_PARSE if the headers were incomplete.
1553 *********************************************************************/
1554 static jb_err get_server_headers(struct client_state *csp)
1556 int continue_hack_in_da_house = 0;
1559 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1564 * continue hack in da house. Ignore the ending of
1565 * this head and continue enlisting header lines.
1566 * The reason is described below.
1568 enlist(csp->headers, "");
1569 continue_hack_in_da_house = 0;
1572 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1575 * It's a bodyless continue response, don't
1576 * stop header parsing after reaching its end.
1578 * As a result Privoxy will concatenate the
1579 * next response's head and parse and deliver
1580 * the headers as if they belonged to one request.
1582 * The client will separate them because of the
1583 * empty line between them.
1585 * XXX: What we're doing here is clearly against
1586 * the intended purpose of the continue header,
1587 * and under some conditions (HTTP/1.0 client request)
1588 * it's a standard violation.
1590 * Anyway, "sort of against the spec" is preferable
1591 * to "always getting confused by Continue responses"
1592 * (Privoxy's behaviour before this hack was added)
1594 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1595 continue_hack_in_da_house = 1;
1597 else if (*header == '\0')
1600 * If the header is empty, but the Continue hack
1601 * isn't active, we can assume that we reached the
1602 * end of the buffer before we hit the end of the
1605 * Inform the caller an let it decide how to handle it.
1607 return JB_ERR_PARSE;
1610 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1613 * XXX: Should we quit the request and return a
1614 * out of memory error page instead?
1616 log_error(LOG_LEVEL_ERROR,
1617 "Out of memory while enlisting server headers. %s lost.",
1627 /*********************************************************************
1629 * Function : crunch_reason
1631 * Description : Translates the crunch reason code into a string.
1634 * 1 : rsp = a http_response
1636 * Returns : A string with the crunch reason or an error description.
1638 *********************************************************************/
1639 static const char *crunch_reason(const struct http_response *rsp)
1641 char * reason = NULL;
1643 assert(rsp != NULL);
1646 return "Internal error while searching for crunch reason";
1649 switch (rsp->reason)
1651 case RSP_REASON_UNSUPPORTED:
1652 reason = "Unsupported HTTP feature";
1654 case RSP_REASON_BLOCKED:
1657 case RSP_REASON_UNTRUSTED:
1658 reason = "Untrusted";
1660 case RSP_REASON_REDIRECTED:
1661 reason = "Redirected";
1663 case RSP_REASON_CGI_CALL:
1664 reason = "CGI Call";
1666 case RSP_REASON_NO_SUCH_DOMAIN:
1667 reason = "DNS failure";
1669 case RSP_REASON_FORWARDING_FAILED:
1670 reason = "Forwarding failed";
1672 case RSP_REASON_CONNECT_FAILED:
1673 reason = "Connection failure";
1675 case RSP_REASON_OUT_OF_MEMORY:
1676 reason = "Out of memory (may mask other reasons)";
1679 reason = "No reason recorded";
1687 /*********************************************************************
1689 * Function : send_crunch_response
1691 * Description : Delivers already prepared response for
1692 * intercepted requests, logs the interception
1693 * and frees the response.
1696 * 1 : csp = Current client state (buffers, headers, etc...)
1697 * 1 : rsp = Fully prepared response. Will be freed on exit.
1699 * Returns : Nothing.
1701 *********************************************************************/
1702 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1704 const struct http_request *http = csp->http;
1705 char status_code[4];
1707 assert(rsp != NULL);
1708 assert(rsp->head != NULL);
1713 * Not supposed to happen. If it does
1714 * anyway, treat it as an unknown error.
1716 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1717 /* return code doesn't matter */
1722 /* If rsp is still NULL, we have serious internal problems. */
1723 log_error(LOG_LEVEL_FATAL,
1724 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1728 * Extract the status code from the actual head
1729 * that was send to the client. It is the only
1730 * way to get it right for all requests, including
1731 * the fixed ones for out-of-memory problems.
1733 * A head starts like this: 'HTTP/1.1 200...'
1737 status_code[0] = rsp->head[9];
1738 status_code[1] = rsp->head[10];
1739 status_code[2] = rsp->head[11];
1740 status_code[3] = '\0';
1742 /* Write the answer to the client */
1743 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1744 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1746 /* There is nothing we can do about it. */
1747 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1750 /* Log that the request was crunched and why. */
1751 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1752 http->hostport, http->path, crunch_reason(rsp));
1753 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1754 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1756 /* Clean up and return */
1757 if (cgi_error_memory() != rsp)
1759 free_http_response(rsp);
1766 /*********************************************************************
1768 * Function : request_contains_null_bytes
1770 * Description : Checks for NULL bytes in the request and sends
1771 * an error message to the client if any were found.
1773 * XXX: currently not used, see comment in chat().
1776 * 1 : csp = Current client state (buffers, headers, etc...)
1777 * 2 : buf = Data from the client's request to check.
1778 * 3 : len = The data length.
1780 * Returns : TRUE if the request contained one or more NULL bytes, or
1783 *********************************************************************/
1784 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1786 size_t c_len; /* Request lenght when treated as C string */
1788 c_len = strlen(buf);
1793 * Null byte(s) found. Log the request,
1794 * return an error response and hang up.
1796 size_t tmp_len = c_len;
1801 * Replace NULL byte(s) with '°' characters
1802 * so the request can be logged as string.
1803 * XXX: Is there a better replacement character?
1806 tmp_len += strlen(buf+tmp_len);
1807 } while (tmp_len < len);
1809 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1810 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1811 log_error(LOG_LEVEL_HEADER,
1812 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1814 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1816 /* XXX: Log correct size */
1817 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1827 /*********************************************************************
1829 * Function : crunch_response_triggered
1831 * Description : Checks if the request has to be crunched,
1832 * and delivers the crunch response if necessary.
1835 * 1 : csp = Current client state (buffers, headers, etc...)
1836 * 2 : crunchers = list of cruncher functions to run
1838 * Returns : TRUE if the request was answered with a crunch response
1841 *********************************************************************/
1842 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1844 struct http_response *rsp = NULL;
1845 const struct cruncher *c;
1848 * If CGI request crunching is disabled,
1849 * check the CGI dispatcher out of order to
1850 * prevent unintentional blocks or redirects.
1852 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1853 && (NULL != (rsp = dispatch_cgi(csp))))
1855 /* Deliver, log and free the interception response. */
1856 send_crunch_response(csp, rsp);
1860 for (c = crunchers; c->cruncher != NULL; c++)
1863 * Check the cruncher if either Privoxy is toggled
1864 * on and the request isn't forced, or if the cruncher
1865 * applies to forced requests as well.
1867 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1868 !(csp->flags & CSP_FLAG_FORCED)) ||
1869 (c->flags & CF_IGNORE_FORCE))
1871 rsp = c->cruncher(csp);
1874 /* Deliver, log and free the interception response. */
1875 send_crunch_response(csp, rsp);
1876 #ifdef FEATURE_STATISTICS
1877 if (c->flags & CF_COUNT_AS_REJECT)
1879 csp->flags |= CSP_FLAG_REJECTED;
1881 #endif /* def FEATURE_STATISTICS */
1892 /*********************************************************************
1894 * Function : build_request_line
1896 * Description : Builds the HTTP request line.
1898 * If a HTTP forwarder is used it expects the whole URL,
1899 * web servers only get the path.
1902 * 1 : csp = Current client state (buffers, headers, etc...)
1903 * 2 : fwd = The forwarding spec used for the request
1904 * XXX: Should use http->fwd instead.
1905 * 3 : request_line = The old request line which will be replaced.
1907 * Returns : Nothing. Terminates in case of memory problems.
1909 *********************************************************************/
1910 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1912 struct http_request *http = csp->http;
1914 assert(http->ssl == 0);
1917 * Downgrade http version from 1.1 to 1.0
1918 * if +downgrade action applies.
1920 if ( (csp->action->flags & ACTION_DOWNGRADE)
1921 && (!strcmpic(http->ver, "HTTP/1.1")))
1924 http->ver = strdup("HTTP/1.0");
1926 if (http->ver == NULL)
1928 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1933 * Rebuild the request line.
1935 freez(*request_line);
1936 *request_line = strdup(http->gpc);
1937 string_append(request_line, " ");
1939 if (fwd->forward_host)
1941 string_append(request_line, http->url);
1945 string_append(request_line, http->path);
1947 string_append(request_line, " ");
1948 string_append(request_line, http->ver);
1950 if (*request_line == NULL)
1952 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1954 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1958 /*********************************************************************
1960 * Function : change_request_destination
1962 * Description : Parse a (rewritten) request line and regenerate
1963 * the http request data.
1966 * 1 : csp = Current client state (buffers, headers, etc...)
1968 * Returns : Forwards the parse_http_request() return code.
1969 * Terminates in case of memory problems.
1971 *********************************************************************/
1972 static jb_err change_request_destination(struct client_state *csp)
1974 struct http_request *http = csp->http;
1977 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1978 free_http_request(http);
1979 err = parse_http_request(csp->headers->first->str, http, csp);
1980 if (JB_ERR_OK != err)
1982 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1983 jb_err_to_string(err));
1987 /* XXX: ocmd is a misleading name */
1988 http->ocmd = strdup(http->cmd);
1989 if (http->ocmd == NULL)
1991 log_error(LOG_LEVEL_FATAL,
1992 "Out of memory copying rewritten HTTP request line");
2000 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2001 /*********************************************************************
2003 * Function : server_response_is_complete
2005 * Description : Determines whether we should stop reading
2006 * from the server socket.
2009 * 1 : csp = Current client state (buffers, headers, etc...)
2010 * 2 : content_length = Length of content received so far.
2012 * Returns : TRUE if the response is complete,
2015 *********************************************************************/
2016 static int server_response_is_complete(struct client_state *csp, size_t content_length)
2018 int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2020 if (!strcmpic(csp->http->gpc, "HEAD"))
2023 * "HEAD" implies no body, we are thus expecting
2024 * no content. XXX: incomplete "list" of methods?
2026 log_error(LOG_LEVEL_INFO, "Method %s implies no body.", csp->http->gpc);
2027 csp->expected_content_length = 0;
2028 content_length_known = TRUE;
2031 if (csp->http->status == 304)
2034 * Expect no body. XXX: incomplete "list" of status codes?
2036 log_error(LOG_LEVEL_INFO, "Status code %d implies no body.", csp->http->status);
2037 csp->expected_content_length = 0;
2038 content_length_known = TRUE;
2041 return (content_length_known && ((0 == csp->expected_content_length)
2042 || (csp->expected_content_length <= content_length)));
2044 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2047 /*********************************************************************
2051 * Description : Once a connection to the client has been accepted,
2052 * this function is called (via serve()) to handle the
2053 * main business of the communication. When this
2054 * function returns, the caller must close the client
2057 * FIXME: chat is nearly thousand lines long.
2061 * 1 : csp = Current client state (buffers, headers, etc...)
2063 * Returns : Nothing.
2065 *********************************************************************/
2066 static void chat(struct client_state *csp)
2068 char buf[BUFFER_SIZE];
2076 int ms_iis5_hack = 0;
2077 size_t byte_count = 0;
2078 int forwarded_connect_retries = 0;
2079 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2080 const struct forward_spec * fwd;
2081 struct http_request *http;
2082 int len; /* for buffer sizes (and negative error codes) */
2085 /* Function that does the content filtering for the current request */
2086 filter_function_ptr content_filter = NULL;
2088 /* Skeleton for HTTP response, if we should intercept the request */
2089 struct http_response *rsp;
2091 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2092 struct list header_list;
2093 struct list *headers = &header_list;
2097 memset(buf, 0, sizeof(buf));
2100 * Read the client's request. Note that since we're not using select() we
2101 * could get blocked here if a client connected, then didn't say anything!
2106 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2108 if (len <= 0) break; /* error! */
2111 * If there is no memory left for buffering the
2112 * request, there is nothing we can do but hang up
2114 if (add_to_iob(csp, buf, len))
2119 req = get_header(csp->iob);
2121 } while ((NULL != req) && ('\0' == *req));
2123 if ((NULL != req) && ('\0' != *req))
2125 /* Request received. Validate and parse it. */
2129 * XXX: Temporary disabled to prevent problems
2130 * with POST requests whose bodies are allowed to
2131 * contain NULL bytes. BR#1730105.
2133 * The main purpose of this check is to properly
2134 * log stuff like BitTorrent traffic and other junk
2135 * that hits public proxies. It's not required for
2136 * Privoxy to functions as those requests are discarded
2139 * It probably should be rewritten to only check
2140 * the head of the request. Another option would
2141 * be to let all POST requests pass, although that
2142 * may not be good enough.
2144 if (request_contains_null_bytes(csp, buf, len))
2146 /* NULL bytes found and dealt with, just hang up. */
2151 /* Does the request line look invalid? */
2152 if (client_protocol_is_unsupported(csp, req))
2155 * Yes. The request has already been
2156 * answered with a error response, the buffers
2157 * were freed and we're done with chatting.
2162 #ifdef FEATURE_FORCE_LOAD
2164 * If this request contains the FORCE_PREFIX and blocks
2165 * aren't enforced, get rid of it and set the force flag.
2167 if (strstr(req, FORCE_PREFIX))
2169 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2171 log_error(LOG_LEVEL_FORCE,
2172 "Ignored force prefix in request: \"%s\".", req);
2176 strclean(req, FORCE_PREFIX);
2177 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2178 csp->flags |= CSP_FLAG_FORCED;
2182 #endif /* def FEATURE_FORCE_LOAD */
2183 err = parse_http_request(req, http, csp);
2184 if (JB_ERR_OK != err)
2186 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2192 if (http->cmd == NULL)
2194 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2195 /* XXX: Use correct size */
2196 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2197 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2199 free_http_request(http);
2203 /* grab the rest of the client's headers */
2207 p = get_header(csp->iob);
2211 /* There are no additional headers to read. */
2218 * We didn't receive a complete header
2219 * line yet, get the rest of it.
2221 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2224 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2225 destroy_list(headers);
2229 if (add_to_iob(csp, buf, len))
2232 * If there is no memory left for buffering the
2233 * request, there is nothing we can do but hang up
2235 destroy_list(headers);
2242 * We were able to read a complete
2243 * header and can finaly enlist it.
2250 if (http->host == NULL)
2253 * If we still don't know the request destination,
2254 * the request is invalid or the client uses
2255 * Privoxy without its knowledge.
2257 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2260 * Our attempts to get the request destination
2261 * elsewhere failed or Privoxy is configured
2262 * to only accept proxy requests.
2264 * An error response has already been send
2265 * and we're done here.
2272 * Determine the actions for this URL
2274 #ifdef FEATURE_TOGGLE
2275 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2277 /* Most compatible set of actions (i.e. none) */
2278 init_current_action(csp->action);
2281 #endif /* ndef FEATURE_TOGGLE */
2283 get_url_actions(csp, http);
2287 * Save a copy of the original request for logging
2289 http->ocmd = strdup(http->cmd);
2291 if (http->ocmd == NULL)
2293 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2296 enlist(csp->headers, http->cmd);
2298 /* Append the previously read headers */
2299 list_append_list_unique(csp->headers, headers);
2300 destroy_list(headers);
2302 err = sed(csp, FILTER_CLIENT_HEADERS);
2303 if (JB_ERR_OK != err)
2305 assert(err == JB_ERR_PARSE);
2306 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2308 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2310 /* Check request line for rewrites. */
2311 if ((NULL == csp->headers->first->str)
2312 || (strcmp(http->cmd, csp->headers->first->str) &&
2313 (JB_ERR_OK != change_request_destination(csp))))
2316 * A header filter broke the request line - bail out.
2318 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2319 /* XXX: Use correct size */
2320 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2321 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2323 free_http_request(http);
2327 /* decide how to route the HTTP request */
2328 fwd = forward_url(csp, http);
2331 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2332 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2336 * build the http request to send to the server
2337 * we have to do one of the following:
2339 * create = use the original HTTP request to create a new
2340 * HTTP request that has either the path component
2341 * without the http://domainspec (w/path) or the
2342 * full orininal URL (w/url)
2343 * Note that the path and/or the HTTP version may
2344 * have been altered by now.
2346 * connect = Open a socket to the host:port of the server
2347 * and short-circuit server and client socket.
2349 * pass = Pass the request unchanged if forwarding a CONNECT
2350 * request to a parent proxy. Note that we'll be sending
2351 * the CFAIL message ourselves if connecting to the parent
2352 * fails, but we won't send a CSUCCEED message if it works,
2353 * since that would result in a double message (ours and the
2354 * parent's). After sending the request to the parent, we simply
2357 * here's the matrix:
2360 * +--------+--------+
2362 * 0 | create | connect|
2364 * Forwarding +--------+--------+
2366 * 1 | create | pass |
2368 * +--------+--------+
2372 if (http->ssl && connect_port_is_forbidden(csp))
2374 const char *acceptable_connect_ports =
2375 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2376 assert(NULL != acceptable_connect_ports);
2377 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2378 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2379 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2380 csp->action->flags |= ACTION_BLOCK;
2386 freez(csp->headers->first->str);
2387 build_request_line(csp, fwd, &csp->headers->first->str);
2390 hdr = list_to_text(csp->headers);
2393 /* FIXME Should handle error properly */
2394 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2398 * We have a request. Check if one of the crunchers wants it.
2400 if (crunch_response_triggered(csp, crunchers_all))
2403 * Yes. The client got the crunch response
2404 * and we are done here after cleaning up.
2407 list_remove_all(csp->headers);
2413 * The headers can't be removed earlier because
2414 * they were still needed for the referrer check
2415 * in case of CGI crunches.
2417 * XXX: Would it be worth to move the referrer check
2418 * into client_referrer() and set a flag if it's trusted?
2420 list_remove_all(csp->headers);
2422 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2424 if (fwd->forward_host)
2426 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2427 fwd->forward_host, fwd->forward_port, http->hostport);
2431 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2434 /* here we connect to the server, gateway, or the forwarder */
2436 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2437 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2439 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2440 forwarded_connect_retries, http->hostport);
2443 if (csp->sfd == JB_INVALID_SOCKET)
2445 if (fwd->type != SOCKS_NONE)
2448 rsp = error_response(csp, "forwarding-failed", errno);
2450 else if (errno == EINVAL)
2452 rsp = error_response(csp, "no-such-domain", errno);
2456 rsp = error_response(csp, "connect-failed", errno);
2457 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2461 /* Write the answer to the client */
2464 send_crunch_response(csp, rsp);
2471 if (fwd->forward_host || (http->ssl == 0))
2474 * Write the client's (modified) header to the server
2475 * (along with anything else that may be in the buffer)
2477 if (write_socket(csp->sfd, hdr, strlen(hdr))
2478 || (flush_socket(csp->sfd, csp->iob) < 0))
2480 log_error(LOG_LEVEL_CONNECT,
2481 "write header to: %s failed: %E", http->hostport);
2483 rsp = error_response(csp, "connect-failed", errno);
2486 send_crunch_response(csp, rsp);
2496 * We're running an SSL tunnel and we're not forwarding,
2497 * so just send the "connect succeeded" message to the
2498 * client, flush the rest, and get out of the way.
2500 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2508 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2510 /* we're finished with the client's header */
2513 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2515 /* pass data between the client and server
2516 * until one or the other shuts down the connection.
2525 * FD_ZERO here seems to point to an errant macro which crashes.
2526 * So do this by hand for now...
2528 memset(&rfds,0x00,sizeof(fd_set));
2532 FD_SET(csp->cfd, &rfds);
2533 FD_SET(csp->sfd, &rfds);
2535 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2536 if (server_body && server_response_is_complete(csp, byte_count))
2538 log_error(LOG_LEVEL_CONNECT,
2539 "Done reading from server. Expected content length: %d. "
2540 "Actual content length: %d. Most recently received: %d.",
2541 csp->expected_content_length, byte_count, len);
2544 * XXX: should not jump around,
2545 * chat() is complicated enough already.
2549 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2551 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2555 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2560 * This is the body of the browser's request,
2561 * just read and write it.
2563 if (FD_ISSET(csp->cfd, &rfds))
2565 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2569 break; /* "game over, man" */
2572 if (write_socket(csp->sfd, buf, (size_t)len))
2574 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2581 * The server wants to talk. It could be the header or the body.
2582 * If `hdr' is null, then it's the header otherwise it's the body.
2583 * FIXME: Does `hdr' really mean `host'? No.
2585 if (FD_ISSET(csp->sfd, &rfds))
2588 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2592 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2594 if (http->ssl && (fwd->forward_host == NULL))
2597 * Just hang up. We already confirmed the client's CONNECT
2598 * request with status code 200 and unencrypted content is
2599 * no longer welcome.
2601 log_error(LOG_LEVEL_ERROR,
2602 "CONNECT already confirmed. Unable to tell the client about the problem.");
2605 else if (byte_count)
2608 * Just hang up. We already transmitted the original headers
2609 * and parts of the original content and therefore missed the
2610 * chance to send an error message (without risking data corruption).
2612 * XXX: we could retry with a fancy range request here.
2614 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2615 "Unable to tell the client about the problem.");
2619 rsp = error_response(csp, "connect-failed", errno);
2622 send_crunch_response(csp, rsp);
2628 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2629 if (csp->flags & CSP_FLAG_CHUNKED)
2631 if ((len > 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
2633 /* XXX: this is a temporary hack */
2634 log_error(LOG_LEVEL_CONNECT,
2635 "Looks like we reached the end of the last chunk: "
2636 "%d %d %d %d %d. We better stop reading.",
2637 buf[len-5], buf[len-4], buf[len-3], buf[len-2], buf[len-1]);
2638 csp->expected_content_length = byte_count + len;
2639 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2643 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2646 * Add a trailing zero to let be able to use string operations.
2647 * XXX: do we still need this with filter_popups gone?
2652 * Normally, this would indicate that we've read
2653 * as much as the server has sent us and we can
2654 * close the client connection. However, Microsoft
2655 * in its wisdom has released IIS/5 with a bug that
2656 * prevents it from sending the trailing \r\n in
2657 * a 302 redirect header (and possibly other headers).
2658 * To work around this if we've haven't parsed
2659 * a full header we'll append a trailing \r\n
2660 * and see if this now generates a valid one.
2662 * This hack shouldn't have any impacts. If we've
2663 * already transmitted the header or if this is a
2664 * SSL connection, then we won't bother with this
2665 * hack. So we only work on partially received
2666 * headers. If we append a \r\n and this still
2667 * doesn't generate a valid header, then we won't
2668 * transmit anything to the client.
2673 if (server_body || http->ssl)
2676 * If we have been buffering up the document,
2677 * now is the time to apply content modification
2678 * and send the result to the client.
2682 p = execute_content_filter(csp, content_filter);
2684 * If the content filter fails, use the original
2685 * buffer and length.
2686 * (see p != NULL ? p : csp->iob->cur below)
2690 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2693 if (JB_ERR_OK != update_server_headers(csp))
2695 log_error(LOG_LEVEL_FATAL,
2696 "Failed to update server headers. after filtering.");
2699 hdr = list_to_text(csp->headers);
2702 /* FIXME Should handle error properly */
2703 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2706 if (write_socket(csp->cfd, hdr, strlen(hdr))
2707 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2709 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2719 break; /* "game over, man" */
2723 * This is NOT the body, so
2724 * Let's pretend the server just sent us a blank line.
2726 snprintf(buf, sizeof(buf), "\r\n");
2727 len = (int)strlen(buf);
2730 * Now, let the normal header parsing algorithm below do its
2731 * job. If it fails, we'll exit instead of continuing.
2738 * If this is an SSL connection or we're in the body
2739 * of the server document, just write it to the client,
2740 * unless we need to buffer the body for later content-filtering
2742 if (server_body || http->ssl)
2747 * If there is no memory left for buffering the content, or the buffer limit
2748 * has been reached, switch to non-filtering mode, i.e. make & write the
2749 * header, flush the iob and buf, and get out of the way.
2751 if (add_to_iob(csp, buf, len))
2756 log_error(LOG_LEVEL_INFO,
2757 "Flushing header and buffers. Stepping back from filtering.");
2759 hdr = list_to_text(csp->headers);
2763 * Memory is too tight to even generate the header.
2764 * Send our static "Out-of-memory" page.
2766 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2767 rsp = cgi_error_memory();
2768 send_crunch_response(csp, rsp);
2772 hdrlen = strlen(hdr);
2774 if (write_socket(csp->cfd, hdr, hdrlen)
2775 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2776 || (write_socket(csp->cfd, buf, (size_t)len)))
2778 log_error(LOG_LEVEL_CONNECT,
2779 "Flush header and buffers to client failed: %E");
2785 * Reset the byte_count to the amount of bytes
2786 * we just flushed. len will be added a few lines below,
2787 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2789 byte_count = (size_t)flushed;
2791 content_filter = NULL;
2797 if (write_socket(csp->cfd, buf, (size_t)len))
2799 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2803 byte_count += (size_t)len;
2808 const char *header_start;
2810 * We're still looking for the end of the server's header.
2811 * Buffer up the data we just read. If that fails, there's
2812 * little we can do but send our static out-of-memory page.
2814 if (add_to_iob(csp, buf, len))
2816 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2817 rsp = cgi_error_memory();
2818 send_crunch_response(csp, rsp);
2823 header_start = csp->iob->cur;
2825 /* Convert iob into something sed() can digest */
2826 if (JB_ERR_PARSE == get_server_headers(csp))
2831 * Well, we tried our MS IIS/5 hack and it didn't work.
2832 * The header is incomplete and there isn't anything
2833 * we can do about it.
2835 log_error(LOG_LEVEL_INFO,
2836 "MS IIS5 hack didn't produce valid headers.");
2842 * Since we have to wait for more from the server before
2843 * we can parse the headers we just continue here.
2849 /* Did we actually get anything? */
2850 if (NULL == csp->headers->first)
2852 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2853 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2854 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2855 free_http_request(http);
2859 assert(csp->headers->first->str);
2861 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2862 strncmpic(csp->headers->first->str, "ICY", 3))
2865 * It doesn't look like a HTTP (or Shoutcast) response:
2866 * tell the client and log the problem.
2868 if (strlen(csp->headers->first->str) > 30)
2870 csp->headers->first->str[30] = '\0';
2872 log_error(LOG_LEVEL_ERROR,
2873 "Invalid server or forwarder response. Starts with: %s",
2874 csp->headers->first->str);
2875 log_error(LOG_LEVEL_CLF,
2876 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2877 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2878 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2879 free_http_request(http);
2884 * We have now received the entire server header,
2885 * filter it and send the result to the client
2887 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2889 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2891 hdr = list_to_text(csp->headers);
2894 /* FIXME Should handle error properly */
2895 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2898 if (crunch_response_triggered(csp, crunchers_light))
2901 * One of the tags created by a server-header
2902 * tagger triggered a crunch. We already
2903 * delivered the crunch response to the client
2904 * and are done here after cleaning up.
2909 /* Buffer and pcrs filter this if appropriate. */
2911 if (!http->ssl) /* We talk plaintext */
2913 content_filter = get_filter_function(csp);
2916 * Only write if we're not buffering for content modification
2918 if (!content_filter)
2921 * Write the server's (modified) header to
2922 * the client (along with anything else that
2923 * may be in the buffer)
2926 if (write_socket(csp->cfd, hdr, strlen(hdr))
2927 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2929 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2932 * The write failed, so don't bother mentioning it
2933 * to the client... it probably can't hear us anyway.
2939 byte_count += (size_t)len;
2944 * XXX: the header lenght should probably
2945 * be calculated by get_server_headers().
2947 int header_length = csp->iob->cur - header_start;
2948 assert(csp->iob->cur > header_start);
2949 byte_count += len - header_length;
2952 /* we're finished with the server's header */
2958 * If this was a MS IIS/5 hack then it means the server
2959 * has already closed the connection. Nothing more to read.
2964 log_error(LOG_LEVEL_INFO,
2965 "Closed server connection detected with MS IIS5 hack enabled.");
2972 return; /* huh? we should never get here */
2975 if (csp->content_length == 0)
2978 * If Privoxy didn't recalculate the Content-Lenght,
2979 * byte_count is still correct.
2981 csp->content_length = byte_count;
2984 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2985 csp->ip_addr_str, http->ocmd, csp->content_length);
2989 /*********************************************************************
2993 * Description : This is little more than chat. We only "serve" to
2994 * to close any socket that chat may have opened.
2997 * 1 : csp = Current client state (buffers, headers, etc...)
3001 *********************************************************************/
3003 void serve(struct client_state *csp)
3004 #else /* ifndef AMIGA */
3005 static void serve(struct client_state *csp)
3006 #endif /* def AMIGA */
3009 close_socket(csp->cfd);
3011 if (csp->sfd != JB_INVALID_SOCKET)
3013 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3014 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
3016 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
3020 forget_connection(csp->sfd);
3021 close_socket(csp->sfd);
3024 close_socket(csp->sfd);
3025 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3028 csp->flags &= ~CSP_FLAG_ACTIVE;
3034 /*********************************************************************
3036 * Function : server_thread
3038 * Description : We only exist to call `serve' in a threaded environment.
3041 * 1 : data = Current client state (buffers, headers, etc...)
3043 * Returns : Always 0.
3045 *********************************************************************/
3046 static int32 server_thread(void *data)
3048 serve((struct client_state *) data);
3055 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3056 /*********************************************************************
3060 * Description : Print usage info & exit.
3062 * Parameters : Pointer to argv[0] for identifying ourselves
3066 *********************************************************************/
3067 static void usage(const char *myname)
3069 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3073 #endif /* defined(unix) */
3076 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3077 #endif /* defined(unix) */
3078 "[--version] [configfile]\n"
3079 "Aborting\n", myname);
3084 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3087 #ifdef MUTEX_LOCKS_AVAILABLE
3088 /*********************************************************************
3090 * Function : privoxy_mutex_lock
3092 * Description : Locks a mutex.
3095 * 1 : mutex = The mutex to lock.
3097 * Returns : Void. May exit in case of errors.
3099 *********************************************************************/
3100 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3102 #ifdef FEATURE_PTHREAD
3103 int err = pthread_mutex_lock(mutex);
3106 if (mutex != &log_mutex)
3108 log_error(LOG_LEVEL_FATAL,
3109 "Mutex locking failed: %s.\n", strerror(err));
3114 EnterCriticalSection(mutex);
3115 #endif /* def FEATURE_PTHREAD */
3119 /*********************************************************************
3121 * Function : privoxy_mutex_unlock
3123 * Description : Unlocks a mutex.
3126 * 1 : mutex = The mutex to unlock.
3128 * Returns : Void. May exit in case of errors.
3130 *********************************************************************/
3131 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3133 #ifdef FEATURE_PTHREAD
3134 int err = pthread_mutex_unlock(mutex);
3137 if (mutex != &log_mutex)
3139 log_error(LOG_LEVEL_FATAL,
3140 "Mutex unlocking failed: %s.\n", strerror(err));
3145 LeaveCriticalSection(mutex);
3146 #endif /* def FEATURE_PTHREAD */
3150 /*********************************************************************
3152 * Function : privoxy_mutex_init
3154 * Description : Prepares a mutex.
3157 * 1 : mutex = The mutex to initialize.
3159 * Returns : Void. May exit in case of errors.
3161 *********************************************************************/
3162 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3164 #ifdef FEATURE_PTHREAD
3165 int err = pthread_mutex_init(mutex, 0);
3168 printf("Fatal error. Mutex initialization failed: %s.\n",
3173 InitializeCriticalSection(mutex);
3174 #endif /* def FEATURE_PTHREAD */
3176 #endif /* def MUTEX_LOCKS_AVAILABLE */
3178 /*********************************************************************
3180 * Function : initialize_mutexes
3182 * Description : Prepares mutexes if mutex support is available.
3186 * Returns : Void, exits in case of errors.
3188 *********************************************************************/
3189 static void initialize_mutexes(void)
3191 #ifdef MUTEX_LOCKS_AVAILABLE
3193 * Prepare global mutex semaphores
3195 privoxy_mutex_init(&log_mutex);
3196 privoxy_mutex_init(&log_init_mutex);
3197 privoxy_mutex_init(&connection_reuse_mutex);
3200 * XXX: The assumptions below are a bit naive
3201 * and can cause locks that aren't necessary.
3203 * For example older FreeBSD versions (< 6.x?)
3204 * have no gethostbyname_r, but gethostbyname is
3207 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3208 privoxy_mutex_init(&resolver_mutex);
3209 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3211 * XXX: should we use a single mutex for
3212 * localtime() and gmtime() as well?
3214 #ifndef HAVE_GMTIME_R
3215 privoxy_mutex_init(&gmtime_mutex);
3216 #endif /* ndef HAVE_GMTIME_R */
3218 #ifndef HAVE_LOCALTIME_R
3219 privoxy_mutex_init(&localtime_mutex);
3220 #endif /* ndef HAVE_GMTIME_R */
3223 privoxy_mutex_init(&rand_mutex);
3224 #endif /* ndef HAVE_RANDOM */
3225 #endif /* def MUTEX_LOCKS_AVAILABLE */
3229 /*********************************************************************
3233 * Description : Load the config file and start the listen loop.
3234 * This function is a lot more *sane* with the `load_config'
3235 * and `listen_loop' functions; although it stills does
3236 * a *little* too much for my taste.
3239 * 1 : argc = Number of parameters (including $0).
3240 * 2 : argv = Array of (char *)'s to the parameters.
3242 * Returns : 1 if : can't open config file, unrecognized directive,
3243 * stats requested in multi-thread mode, can't open the
3244 * log file, can't open the jar file, listen port is invalid,
3245 * any load fails, and can't bind port.
3247 * Else main never returns, the process must be signaled
3248 * to terminate execution. Or, on Windows, use the
3249 * "File", "Exit" menu option.
3251 *********************************************************************/
3253 int real_main(int argc, const char *argv[])
3255 int main(int argc, const char *argv[])
3259 unsigned int random_seed;
3261 struct passwd *pw = NULL;
3262 struct group *grp = NULL;
3265 char *pre_chroot_nslookup_to_load_resolver = NULL;
3272 #if !defined(_WIN32)
3280 * Parse the command line arguments
3282 * XXX: simply printing usage information in case of
3283 * invalid arguments isn't particularly user friendly.
3285 while (++argc_pos < argc)
3288 /* Check to see if the service must be installed or uninstalled */
3289 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3291 const char *pName = argv[argc_pos] + 9;
3294 exit( (install_service(pName)) ? 0 : 1 );
3296 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3298 const char *pName = argv[argc_pos] + 11;
3301 exit((uninstall_service(pName)) ? 0 : 1);
3303 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3305 bRunAsService = TRUE;
3306 w32_set_service_cwd();
3307 atexit(w32_service_exit_notify);
3310 #endif /* defined(_WIN32) */
3313 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3315 if (strcmp(argv[argc_pos], "--help") == 0)
3320 else if(strcmp(argv[argc_pos], "--version") == 0)
3322 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3328 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3333 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3335 if (++argc_pos == argc) usage(argv[0]);
3336 pidfile = strdup(argv[argc_pos]);
3339 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3341 if (++argc_pos == argc) usage(argv[argc_pos]);
3343 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3346 if (NULL == (grp = getgrnam(p)))
3348 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3352 if (NULL == (pw = getpwnam(argv[argc_pos])))
3354 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3357 if (p != NULL) *--p = '\0';
3360 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3362 if (++argc_pos == argc) usage(argv[0]);
3363 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3366 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3370 #endif /* defined(unix) */
3372 else if (argc_pos + 1 != argc)
3375 * This is neither the last command line
3376 * option, nor was it recognized before,
3377 * therefore it must be invalid.
3383 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3385 configfile = argv[argc_pos];
3388 } /* -END- while (more arguments) */
3391 if ( *configfile != '/' )
3393 char cwd[BUFFER_SIZE];
3395 size_t abs_file_size;
3397 /* make config-filename absolute here */
3398 if (NULL == getcwd(cwd, sizeof(cwd)))
3400 perror("failed to get current working directory");
3405 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3406 basedir = strdup(cwd);
3408 if (NULL == basedir ||
3409 NULL == (abs_file = malloc(abs_file_size)))
3411 perror("malloc failed");
3414 strlcpy(abs_file, basedir, abs_file_size);
3415 strlcat(abs_file, "/", abs_file_size );
3416 strlcat(abs_file, configfile, abs_file_size);
3417 configfile = abs_file;
3419 #endif /* defined unix */
3423 clients->next = NULL;
3425 /* XXX: factor out initialising after the next stable release. */
3428 #elif defined(_WIN32)
3432 /* Prepare mutexes if supported and necessary. */
3433 initialize_mutexes();
3435 /* Enable logging until further notice. */
3436 init_log_module(Argv[0]);
3438 random_seed = (unsigned int)time(NULL);
3440 srandom(random_seed);
3443 #endif /* ifdef HAVE_RANDOM */
3446 * Unix signal handling
3448 * Catch the abort, interrupt and terminate signals for a graceful exit
3449 * Catch the hangup signal so the errlog can be reopened.
3450 * Ignore the broken pipe signals (FIXME: Why?)
3452 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3455 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3456 const int ignored_signals[] = { SIGPIPE, 0 };
3458 for (idx = 0; catched_signals[idx] != 0; idx++)
3460 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3461 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3463 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3464 #endif /* ifdef sun */
3466 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3470 for (idx = 0; ignored_signals[idx] != 0; idx++)
3472 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3474 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3479 #else /* ifdef _WIN32 */
3480 # ifdef _WIN_CONSOLE
3482 * We *are* in a windows console app.
3483 * Print a verbose messages about FAQ's and such
3485 printf("%s", win32_blurb);
3486 # endif /* def _WIN_CONSOLE */
3487 #endif /* def _WIN32 */
3490 /* Initialize the CGI subsystem */
3491 cgi_init_error_messages();
3494 * If runnig on unix and without the --nodaemon
3495 * option, become a daemon. I.e. fork, detach
3496 * from tty and get process group leadership
3509 if ( pid < 0 ) /* error */
3514 else if ( pid != 0 ) /* parent */
3519 * must check for errors
3520 * child died due to missing files aso
3523 wpid = waitpid( pid, &status, WNOHANG );
3532 /* Should be more portable, but not as well tested */
3537 #else /* ndef __FreeBSD__ */
3539 #endif /* ndef __FreeBSD__ */
3540 fd = open("/dev/tty", O_RDONLY);
3543 /* no error check here */
3544 ioctl( fd, TIOCNOTTY,0 );
3549 * stderr (fd 2) will be closed later on, when the
3550 * log file has been parsed.
3557 } /* -END- if (!no_daemon) */
3560 * As soon as we have written the PID file, we can switch
3561 * to the user and group ID indicated by the --user option
3567 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3569 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3575 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3577 /* Read the time zone file from /etc before doing chroot. */
3579 if (NULL != pre_chroot_nslookup_to_load_resolver
3580 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3582 /* Initialize resolver library. */
3583 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3585 if (chroot(pw->pw_dir) < 0)
3587 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3591 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3594 if (setuid(pw->pw_uid))
3596 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3600 char putenv_dummy[64];
3602 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3603 if (putenv(putenv_dummy) != 0)
3605 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3608 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3609 if (putenv(putenv_dummy) != 0)
3611 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3617 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3620 #endif /* defined unix */
3623 /* This will be FALSE unless the command line specified --service
3627 /* Yup, so now we must attempt to establish a connection
3628 * with the service dispatcher. This will only work if this
3629 * process was launched by the service control manager to
3630 * actually run as a service. If this isn't the case, i've
3631 * known it take around 30 seconds or so for the call to return.
3634 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3635 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3637 /* Service has run, and at this point is now being stopped, so just return */
3642 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3644 /* An error occurred. Usually it's because --service was wrongly specified
3645 * and we were unable to connect to the Service Control Dispatcher because
3646 * it wasn't expecting us and is therefore not listening.
3648 * For now, just continue below to call the listen_loop function.
3651 #endif /* def _WIN32 */
3661 /*********************************************************************
3663 * Function : bind_port_helper
3665 * Description : Bind the listen port. Handles logging, and aborts
3669 * 1 : config = Privoxy configuration. Specifies port
3672 * Returns : Port that was opened.
3674 *********************************************************************/
3675 static jb_socket bind_port_helper(struct configuration_spec * config)
3680 if (config->haddr == NULL)
3682 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3687 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3688 config->hport, config->haddr);
3691 result = bind_port(config->haddr, config->hport, &bfd);
3698 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3699 "There may be another Privoxy or some other "
3700 "proxy running on port %d",
3701 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3702 config->hport, config->hport);
3705 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3706 "The hostname is not resolvable",
3707 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3710 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3711 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3714 /* shouldn't get here */
3715 return JB_INVALID_SOCKET;
3718 config->need_bind = 0;
3725 /* Without this simple workaround we get this compiler warning from _beginthread
3726 * warning C4028: formal parameter 1 different from declaration
3728 void w32_service_listen_loop(void *p)
3732 #endif /* def _WIN32 */
3735 /*********************************************************************
3737 * Function : listen_loop
3739 * Description : bind the listen port and enter a "FOREVER" listening loop.
3745 *********************************************************************/
3746 static void listen_loop(void)
3748 struct client_state *csp = NULL;
3750 struct configuration_spec * config;
3752 config = load_config();
3754 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3756 * XXX: Should be relocated once it no
3757 * longer needs to emit log messages.
3759 initialize_reusable_connections();
3760 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3762 bfd = bind_port_helper(config);
3764 #ifdef FEATURE_GRACEFUL_TERMINATION
3765 while (!g_terminate)
3770 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3771 while (waitpid(-1, NULL, WNOHANG) > 0)
3773 /* zombie children */
3775 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3778 * Free data that was used by died threads
3784 * Re-open the errlog after HUP signal
3786 if (received_hup_signal)
3788 init_error_log(Argv[0], config->logfile);
3789 received_hup_signal = 0;
3793 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3795 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3799 csp->flags |= CSP_FLAG_ACTIVE;
3800 csp->sfd = JB_INVALID_SOCKET;
3802 csp->config = config = load_config();
3804 if ( config->need_bind )
3807 * Since we were listening to the "old port", we will not see
3808 * a "listen" param change until the next IJB request. So, at
3809 * least 1 more request must be made for us to find the new
3810 * setting. I am simply closing the old socket and binding the
3813 * Which-ever is correct, we will serve 1 more page via the
3814 * old settings. This should probably be a "show-proxy-args"
3815 * request. This should not be a so common of an operation
3816 * that this will hurt people's feelings.
3821 bfd = bind_port_helper(config);
3824 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3826 if (!accept_connection(csp, bfd))
3828 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3841 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3844 #ifdef FEATURE_TOGGLE
3845 if (global_toggle_state)
3846 #endif /* def FEATURE_TOGGLE */
3848 csp->flags |= CSP_FLAG_TOGGLED_ON;
3851 if (run_loader(csp))
3853 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3854 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3858 if (block_acl(NULL,csp))
3860 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3861 close_socket(csp->cfd);
3865 #endif /* def FEATURE_ACL */
3867 /* add it to the list of clients */
3868 csp->next = clients->next;
3869 clients->next = csp;
3871 if (config->multi_threaded)
3875 /* this is a switch () statment in the C preprocessor - ugh */
3876 #undef SELECTED_ONE_OPTION
3878 /* Use Pthreads in preference to native code */
3879 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3880 #define SELECTED_ONE_OPTION
3882 pthread_t the_thread;
3883 pthread_attr_t attrs;
3885 pthread_attr_init(&attrs);
3886 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3887 errno = pthread_create(&the_thread, &attrs,
3888 (void * (*)(void *))serve, csp);
3889 child_id = errno ? -1 : 0;
3890 pthread_attr_destroy(&attrs);
3894 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3895 #define SELECTED_ONE_OPTION
3896 child_id = _beginthread(
3897 (void (*)(void *))serve,
3902 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3903 #define SELECTED_ONE_OPTION
3904 child_id = _beginthread(
3905 (void(* _Optlink)(void*))serve,
3911 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3912 #define SELECTED_ONE_OPTION
3914 thread_id tid = spawn_thread
3915 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3917 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3919 child_id = (int) tid;
3928 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3929 #define SELECTED_ONE_OPTION
3930 csp->cfd = ReleaseSocket(csp->cfd, -1);
3933 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3934 NP_Output, Output(),
3935 NP_CloseOutput, FALSE,
3936 NP_Name, (ULONG)"privoxy child",
3940 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3941 NP_Output, Output(),
3942 NP_CloseOutput, FALSE,
3943 NP_Name, (ULONG)"privoxy child",
3944 NP_StackSize, 200*1024,
3950 ((struct Task *)child_id)->tc_UserData = csp;
3951 Signal((struct Task *)child_id, SIGF_SINGLE);
3956 #if !defined(SELECTED_ONE_OPTION)
3959 /* This block is only needed when using fork().
3960 * When using threads, the server thread was
3961 * created and run by the call to _beginthread().
3963 if (child_id == 0) /* child */
3966 #ifdef FEATURE_TOGGLE
3967 int inherited_toggle_state = global_toggle_state;
3968 #endif /* def FEATURE_TOGGLE */
3973 * If we've been toggled or we've blocked the request, tell Mom
3976 #ifdef FEATURE_TOGGLE
3977 if (inherited_toggle_state != global_toggle_state)
3979 rc |= RC_FLAG_TOGGLED;
3981 #endif /* def FEATURE_TOGGLE */
3983 #ifdef FEATURE_STATISTICS
3984 if (csp->flags & CSP_FLAG_REJECTED)
3986 rc |= RC_FLAG_BLOCKED;
3988 #endif /* ndef FEATURE_STATISTICS */
3992 else if (child_id > 0) /* parent */
3994 /* in a fork()'d environment, the parent's
3995 * copy of the client socket and the CSP
3999 #if !defined(_WIN32) && !defined(__CYGWIN__)
4001 wait( &child_status );
4004 * Evaluate child's return code: If the child has
4005 * - been toggled, toggle ourselves
4006 * - blocked its request, bump up the stats counter
4009 #ifdef FEATURE_TOGGLE
4010 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
4012 global_toggle_state = !global_toggle_state;
4014 #endif /* def FEATURE_TOGGLE */
4016 #ifdef FEATURE_STATISTICS
4018 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4022 #endif /* def FEATURE_STATISTICS */
4024 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4025 close_socket(csp->cfd);
4026 csp->flags &= ~CSP_FLAG_ACTIVE;
4030 #undef SELECTED_ONE_OPTION
4031 /* end of cpp switch () */
4033 if (child_id < 0) /* failed */
4035 char buf[BUFFER_SIZE];
4037 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
4039 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
4041 write_socket(csp->cfd, buf, strlen(buf));
4042 close_socket(csp->cfd);
4043 csp->flags &= ~CSP_FLAG_ACTIVE;
4054 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4056 /* Clean up. Aim: free all memory (no leaks) */
4057 #ifdef FEATURE_GRACEFUL_TERMINATION
4059 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4061 unload_current_config_file();
4062 unload_current_actions_file();
4063 unload_current_re_filterfile();
4064 #ifdef FEATURE_TRUST
4065 unload_current_trust_file();
4068 if (config->multi_threaded)
4075 } while ((clients->next != NULL) && (--i > 0));
4079 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4090 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4091 /* Cleanup - remove taskbar icon etc. */
4096 #endif /* FEATURE_GRACEFUL_TERMINATION */