1 const char jcc_rcs[] = "$Id: jcc.c,v 1.173 2008/05/06 15:09:00 fabiankeil Exp $";
2 /*********************************************************************
4 * File : $Source: /cvsroot/ijbswa/current/jcc.c,v $
6 * Purpose : Main file. Contains main() method, main loop, and
7 * the main connection-handling function.
9 * Copyright : Written by and Copyright (C) 2001-2008 the SourceForge
10 * Privoxy team. http://www.privoxy.org/
12 * Based on the Internet Junkbuster originally written
13 * by and Copyright (C) 1997 Anonymous Coders and
14 * Junkbusters Corporation. http://www.junkbusters.com
16 * This program is free software; you can redistribute it
17 * and/or modify it under the terms of the GNU General
18 * Public License as published by the Free Software
19 * Foundation; either version 2 of the License, or (at
20 * your option) any later version.
22 * This program is distributed in the hope that it will
23 * be useful, but WITHOUT ANY WARRANTY; without even the
24 * implied warranty of MERCHANTABILITY or FITNESS FOR A
25 * PARTICULAR PURPOSE. See the GNU General Public
26 * License for more details.
28 * The GNU General Public License should be included with
29 * this file. If not, you can view it at
30 * http://www.gnu.org/copyleft/gpl.html
31 * or write to the Free Software Foundation, Inc., 59
32 * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
36 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
37 * Least-effort fix for bug #1821930 (reported by Lee):
38 * If the response doesn't look like HTTP,
39 * tell the client and log the problem.
41 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
42 * Don't pass the whole csp structure to flush_socket()
43 * when it only needs a file descriptor and a buffer.
45 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
46 * Remove kill-popups action.
48 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
49 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
51 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
52 * Remove the treat-forbidden-connects-like-blocks action. We now
53 * use the "blocked" page for forbidden CONNECT requests by default.
55 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
56 * Also use shiny new connect_port_is_forbidden() in jcc.c.
58 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
59 * Rename url_actions() to get_url_actions() and let it
60 * use the standard parameter ordering.
62 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
63 * Let forward_url() use the standard parameter ordering
64 * and mark its second parameter immutable.
66 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
67 * Remove the "Listening ... for local connections only" log message.
68 * Whether or not remote connections are able to reach Privoxy is up
69 * to the operating system.
71 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
72 * Prevent the log messages for CONNECT requests to unacceptable
73 * ports from printing the limit-connect argument as [null] if
74 * limit-connect hasn't been explicitly enabled.
76 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
77 * Make sure all console-mode apps get a usage() instance
79 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
80 * Reword NO_SERVER_DATA_RESPONSE to make it harder
81 * to misunderstand what the message is all about.
83 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
84 * Unbreak trustfile which previously didn't work without
85 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
87 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
88 * Plug memory leak. Spotted by Valgrind, triggered by
89 * Privoxy-Regression-Test feeding proxyfuzz.py.
91 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
92 * In the HTTP snipplets, refer to the client as client.
94 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
95 * Emit a log message when activating the MS IIS5 hack.
97 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
98 * Log the "weak randomization factor" warning only
99 * once for mingw32 and provide some more details.
101 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
102 * Initialize log module after initializing mutexes, future
103 * deadlocks in that code should now work cross-platform.
105 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
106 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
107 * Reported by Bert van Leeuwen in BR#1818808.
109 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
110 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
112 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
113 * When in daemon mode, close stderr after the configuration file has been
114 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
116 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
117 * - Fix a crash when parsing invalid requests whose first header
118 * is rejected by get_header(). Regression (re?)introduced
119 * in r1.143 by yours truly.
120 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
121 * client_cookie_adder() to make sure send-vanilla-wafer can be
122 * controlled through tags (and thus regression-tested).
124 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
125 * - Move get_filter_function() from jcc.c to filters.c
126 * so the filter functions can be static.
127 * - Don't bother filtering body-less responses.
129 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
130 * Execute content filters through execute_content_filter().
132 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
133 * Initialize req to NULL to make sure it's defined if the
134 * first read_socket() call fails. Reported by icmp30.
136 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
137 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
138 * extensive comments moved to user manual.
140 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
141 * Don't crash if a broken header filter wiped out the request line.
143 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
144 * Fix byte_count calculation in case of flushes
145 * and don't parse the server headers a second time.
147 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
148 * - If there's a connection problem after we already forwarded
149 * parts of the original content, just hang up. Fixes BR#1776724.
150 * - Fix warnings about unused code on mingw32.
151 * - In case of flushes, calculate the byte count
152 * less incorrectly (I think).
154 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
155 * Add some more prototypes for static functions.
157 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
158 * Comment out request_contains_null_bytes() until it's used again.
160 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
161 * #1763173 from Stefan Huehner: s@const static@static const@
162 * and declare some more functions static.
164 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
165 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
166 * and explain why they were rejected in the first place.
167 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
168 * CONNECT requests. The request line was missing.
169 * - Add two more XXX reminders as we don't have enough already.
171 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
172 * As Hal noticed, checking dispatch_cgi() as the last cruncher
173 * looks like a bug if CGI requests are blocked unintentionally,
174 * so don't do it unless the user enabled the new config option
175 * "allow-cgi-request-crunching".
177 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
178 * - Allow to rewrite the request destination behind the client's back.
179 * - Turn the weird-looking unconditional for loop that
180 * reads the client request into a conditional while loop.
181 * Move the stuff that only runs once out of the loop.
182 * - Move parts of chat(), server_content_type() and the
183 * necessary stuff to fix BR#1750917 into get_filter_function().
185 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
186 * Temporary workaround for BR#1730105.
188 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
189 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
190 * deadlocks and crashes on OpenBSD and possibly other OS with neither
191 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
192 * Thanks to Ralf Horstmann for report and solution.
194 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
195 * Add forward-override{} to change the forwarding settings through
196 * action sections. This is mainly interesting to forward different
197 * clients differently (for example based on User-Agent or request
200 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
201 * - Let usage() mention the --chroot parameter.
202 * - Use read_socket() consistently and always leave
203 * the last buffer byte alone, even in cases where
204 * null termination (currently) doesn't matter.
206 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
207 * - Fix config file loading on Unix if no config file is specified.
208 * Since r1.97 Privoxy would always interpret the last argument as
209 * config file, even if it's a valid command line option.
210 * - Abort in case of unrecognized command line options. Closes #1719696.
211 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
212 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
214 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
215 * - Don't rerun crunchers that only depend on the request URL.
216 * - Don't count redirects and CGI requests as "blocked requests".
218 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
219 * Support crunching based on tags created by server-header taggers.
221 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
222 * Make HTTP snippets static (again). Add a Content-Type for those
223 * with content so the browser doesn't guess it based on the URL.
225 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
226 * Move crunching and request line rebuilding out of chat().
228 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
229 * Introduce tags as alternative way to specify which
230 * actions apply to a request. At the moment tags can be
231 * created based on client and server headers.
233 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
234 * Don't CLF-log CONNECT requests twice.
236 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
237 * Log the source address for ACL-related connection drops.
239 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
240 * New config option: enforce-blocks.
242 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
243 * - Move null byte check into separate function.
244 * - Don't confuse the client with error pages
245 * if a CONNECT request was already confirmed.
247 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
248 * Speed up NULL byte escaping and only log the complete
249 * NULL byte requests with header debugging enabled.
251 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
252 * Answer requests that contain NULL bytes with
253 * a custom response instead of waiting for more
254 * data until the client eventually hangs up.
256 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
257 * - Move delivery and logging of crunched responses
258 * from chat() into send_crunch_response().
259 * - Display the reason for generating http_responses.
260 * - Log the content length for LOG_LEVEL_CLF correctly
261 * (still incorrect for some fixed responses).
262 * - Reword an incorrect comment about
263 * treat-forbidden-connects-like-blocks violating
265 * - Add some log messages.
267 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
268 * Move mutex initialization into separate
269 * function and exit in case of errors.
271 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
272 * - Start to reduce chat()'s line count and move
273 * parts of it into separate functions.
274 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
276 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
277 * - Add Proxy-Agent header to HTTP snippets that are
278 * supposed to reach HTTP clients only.
279 * - Made a few CONNECT log messages more descriptive.
280 * - Catch completely empty server responses (as seen
281 * with Tor's fake ".noconnect" top level domain).
282 * - Use shiny new "forwarding-failed" template for socks errors.
284 * Revision 1.118 2007/01/07 07:43:43 joergs
285 * AmigaOS4 support added.
287 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
288 * Added config option accept-intercepted-requests
289 * and disabled it by default.
291 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
292 * Reverted parts of my last commit
293 * to keep error handling working.
295 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
296 * Fixed gcc43 conversion warnings.
298 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
299 * Fix -pedantic ISO C warning about converting
300 * from function pointer to object pointer.
302 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
303 * Silence compiler warning I introduced with my last commit.
305 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
306 * Mutex protect rand() if POSIX threading
307 * is used, warn the user if that's not possible
308 * and stop using it on _WIN32 where it could
311 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
312 * Don't prevent core dumps by catching SIGABRT.
313 * It's rude and makes debugging unreasonable painful.
315 * Revision 1.110 2006/12/13 14:52:53 etresoft
316 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
318 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
319 * Privoxy is now able to run as intercepting
320 * proxy in combination with any packet filter
321 * that does the port redirection. The destination
322 * is extracted from the "Host:" header which
323 * should be available for nearly all requests.
325 * Moved HTTP snipplets into jcc.c.
326 * Added error message for gopher proxy requests.
328 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
329 * Only unlink the pidfile if it's actually used.
331 * Change order of interception checks to make
332 * it possible to block or redirect requests for
335 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
336 * Make pthread mutex locking more generic. Instead of
337 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
338 * and use mutex locking unless there is an _r function
339 * available. Better safe than sorry.
341 * Fixes "./configure --disable-pthread" and should result
342 * in less threading-related problems on pthread-using platforms,
343 * but it still doesn't fix BR#1122404.
345 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
346 * Move pthread.h inclusion from jcc.c to jcc.h.
347 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
349 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
350 * Don't exit after receiving the second SIGHUP on Solaris.
352 * Fixes BR 1052235, but the same problem may exist on other
353 * systems. Once 3.0.6 is out we should use sigset()
354 * where available and see if it breaks anything.
356 * Revision 1.104 2006/09/23 13:26:38 roro
357 * Replace TABs by spaces in source code.
359 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
360 * Fix +redirect{}. Didn't work with -fast-redirects.
362 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
363 * Respond with 400 and a short text message
364 * if the client tries to use Privoxy as FTP proxy.
366 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
367 * Make number of retries in case of forwarded-connect problems
368 * a config file option (forwarded-connect-retries) and use 0 as
371 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
372 * Set random(3) seed.
374 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
375 * Follow the OpenBSD port's lead and protect the resolve
376 * functions on OpenBSD as well.
378 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
379 * --user fix. Only use the user as group if no group is specified.
380 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
382 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
383 * Windows service (re-)integration
387 * --install[:service_name]
388 * --uninstall[:service_name]
391 * They work as follows:
392 * --install will create a service for you and then terminate.
393 * By default the service name will be "privoxy" (without the quotes).
394 * However you can run multiple services if you wish, just by adding
395 * a colon and then a name (no spaces).
397 * --uninstall follows the exact same rules a --install.
399 * --service is used when the program is executed by the service
400 * control manager, and in normal circumstances would never be
401 * used as a command line argument.
403 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
404 * Windows service integration
406 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
407 * Incorporate Fabian Keil's patch work:
408 http://www.fabiankeil.de/sourcecode/privoxy/
410 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
411 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
412 * with what was really the latest development (the v_3_0_branch branch)
414 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
415 * Thanks to Jindrich Makovicka for a race condition fix for the log
416 * file. The race condition remains for non-pthread implementations.
417 * Reference patch #1175720.
419 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
420 * Add the ability to check jpeg images for invalid
421 * lengths of comment blocks. Defensive strategy
422 * against the exploit:
423 * Microsoft Security Bulletin MS04-028
424 * Buffer Overrun in JPEG Processing (GDI+) Could
425 * Allow Code Execution (833987)
426 * Enabled with +inspect-jpegs in actions files.
428 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
429 * - Fixed usage info for non-unix platforms
430 * - Fixed small cmdline parsing bug
432 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
433 * Diagnostics: Now preserve the returncode of pthread_create
434 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
436 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
437 * No longer ignore SIGCHLD. Fixes bug #769381
439 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
440 * Close jarfile on graceful exit, remove stray line
442 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
443 * Cosmetics: Killed a warning, a typo and an allocation left at exit
445 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
446 * No longer rely on non-POSIX.1 extensions of getcwd().
449 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
450 * Replaced setenv() by posix-compliant putenv()
451 * Thanks to Neil McCalden (nmcc AT users.sf.net).
453 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
454 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
456 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
457 * Clean-up and extension of improvements for forked mode:
458 * - Child's return code now consists of flags RC_FLAG_*
459 * - Reporting toggle to parent now properly #ifdef'ed
460 * - Children now report blocking to parent. This enables
461 * statistics in forked mode
463 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
464 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
465 * Children now report having been toggled through _exit(17), parents
466 * watch for that code and toggle themselves if found.
468 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
469 * Wrapping all *_r functions (the non-_r versions of them) with
470 * mutex semaphores for OSX. Hopefully this will take care of all
471 * of those pesky crash reports.
473 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
474 * Fixed two mostly harmless mem leaks
476 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
477 * Fix: Head of global clients list now initialized to NULL
479 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
480 * Added basic support for OPTIONS and TRACE HTTP methods:
481 * - New interceptor direct_response() added in chat().
482 * - sed() moved to earlier in the process, so that the
483 * Host: header is evaluated before actions and forwarding
486 * Revision 1.92 2002/05/08 16:00:46 oes
487 * Chat's buffer handling:
488 * - Fixed bug with unchecked out-of-mem conditions
489 * while reading client request & server headers
490 * - No longer predict if the buffer limit will be exceeded
491 * in the next read -- check add_to_iob's new
492 * return code. If buffer couldn't be extended
493 * (policy or out-of-mem) while
494 * - reading from client: abort
495 * - reading server headers: send error page
496 * - buffering server body for filter: flush,
497 * and if that fails: send error page
499 * Revision 1.91 2002/04/08 20:35:58 swa
502 * Revision 1.90 2002/04/02 14:57:28 oes
503 * Made sending wafers independent of FEATURE_COOKIE_JAR
505 * Revision 1.89 2002/03/31 17:18:59 jongfoster
506 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
508 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
509 * More compiler warning message maintenance
511 * Revision 1.87 2002/03/26 22:29:54 swa
512 * we have a new homepage!
514 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
515 * Workaround for closing the jarfile before load_config() comes around again
517 * Revision 1.85 2002/03/24 15:23:33 jongfoster
520 * Revision 1.84 2002/03/24 13:25:43 swa
521 * name change related issues
523 * Revision 1.83 2002/03/16 23:54:06 jongfoster
524 * Adding graceful termination feature, to help look for memory leaks.
525 * If you enable this (which, by design, has to be done by hand
526 * editing config.h) and then go to http://i.j.b/die, then the program
527 * will exit cleanly after the *next* request. It should free all the
528 * memory that was used.
530 * Revision 1.82 2002/03/13 00:27:05 jongfoster
533 * Revision 1.81 2002/03/12 01:42:50 oes
534 * Introduced modular filters
536 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
537 * OS/2 port maintenance:
538 * - Fixed EMX build - it had decayed a little
539 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
540 * substituted a memset for now.
542 * Revision 1.79 2002/03/09 20:03:52 jongfoster
543 * - Making various functions return int rather than size_t.
544 * (Undoing a recent change). Since size_t is unsigned on
545 * Windows, functions like read_socket that return -1 on
546 * error cannot return a size_t.
548 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
549 * crashes, and also frequently caused JB to jump to 100%
550 * CPU and stay there. (Because it thought it had just
551 * read ((unsigned)-1) == 4Gb of data...)
553 * - The signature of write_socket has changed, it now simply
554 * returns success=0/failure=nonzero.
556 * - Trying to get rid of a few warnings --with-debug on
557 * Windows, I've introduced a new type "jb_socket". This is
558 * used for the socket file descriptors. On Windows, this
559 * is SOCKET (a typedef for unsigned). Everywhere else, it's
560 * an int. The error value can't be -1 any more, so it's
561 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
562 * Windows it maps to the #define INVALID_SOCKET.)
564 * - The signature of bind_port has changed.
566 * Revision 1.78 2002/03/08 21:35:04 oes
567 * Added optional group supplement to --user option. Will now use default group of user if no group given
569 * Revision 1.77 2002/03/07 03:52:06 oes
570 * - Fixed compiler warnings etc
571 * - Improved handling of failed DNS lookups
573 * Revision 1.76 2002/03/06 22:54:35 jongfoster
574 * Automated function-comment nitpicking.
576 * Revision 1.75 2002/03/06 10:02:19 oes
577 * Fixed stupid bug when --user was not given
579 * Revision 1.74 2002/03/06 00:49:31 jongfoster
580 * Fixing warning on Windows
581 * Making #ifdefs that refer to the same variable consistently
582 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
584 * Revision 1.73 2002/03/05 23:57:30 hal9
585 * Stray character 's' on line 1618 was breaking build.
587 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
588 * - Re-enable OS/2 building after new parms were added
589 * - Fix false out of memory report when resolving CGI templates when no IP
590 * address is available of failed attempt (a la no such domain)
592 * Revision 1.71 2002/03/05 18:13:56 oes
593 * Added --user option
595 * Revision 1.70 2002/03/05 04:52:42 oes
596 * Deleted non-errlog debugging code
598 * Revision 1.69 2002/03/04 23:50:00 jongfoster
599 * Splitting off bind_port() call into bind_port_helper(), with
602 * Revision 1.68 2002/03/04 20:17:32 oes
605 * Revision 1.67 2002/03/04 18:18:57 oes
606 * - Removed _DEBUG mode
607 * - Cleand up cmdline parsing
608 * - Introduced --no-daemon, --pidfile options
609 * - Cleaned up signal handling:
610 * - Terminate cleanly on INT, TERM and ABRT
611 * - Schedule logfile for re-opening on HUP
612 * - Ignore CHLD and PIPE
613 * - Leave the rest with their default handlers
614 * - Uniform handler registration
615 * - Added usage() function
616 * - Played styleguide police
618 * Revision 1.66 2002/03/03 15:06:55 oes
619 * Re-enabled automatic config reloading
621 * Revision 1.65 2002/03/03 14:49:11 oes
622 * Fixed CLF logging: Now uses client's original HTTP request
624 * Revision 1.64 2002/03/03 09:18:03 joergs
625 * Made jumbjuster work on AmigaOS again.
627 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
628 * Clean up a little CRLF unpleasantness that suddenly appeared
630 * Revision 1.62 2002/02/20 23:17:23 jongfoster
631 * Detecting some out-of memory conditions and exiting with a log message.
633 * Revision 1.61 2002/01/17 21:01:52 jongfoster
634 * Moving all our URL and URL pattern parsing code to urlmatch.c.
636 * Revision 1.60 2001/12/30 14:07:32 steudten
637 * - Add signal handling (unix)
638 * - Add SIGHUP handler (unix)
639 * - Add creation of pidfile (unix)
640 * - Add action 'top' in rc file (RH)
641 * - Add entry 'SIGNALS' to manpage
642 * - Add exit message to logfile (unix)
644 * Revision 1.59 2001/12/13 14:07:18 oes
645 * Fixed Bug: 503 error page now sent OK
647 * Revision 1.58 2001/11/30 23:37:24 jongfoster
648 * Renaming the Win32 config file to config.txt - this is almost the
649 * same as the corresponding UNIX name "config"
651 * Revision 1.57 2001/11/16 00:47:43 jongfoster
652 * Changing the tty-disconnection code to use setsid().
654 * Revision 1.56 2001/11/13 20:20:54 jongfoster
655 * Tabs->spaces, fixing a bug with missing {} around an if()
657 * Revision 1.55 2001/11/13 20:14:53 jongfoster
658 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
660 * Revision 1.54 2001/11/07 00:03:14 steudten
661 * Give reliable return value if an error
662 * occurs not just 0 with new daemon mode.
664 * Revision 1.53 2001/11/05 21:41:43 steudten
665 * Add changes to be a real daemon just for unix os.
666 * (change cwd to /, detach from controlling tty, set
667 * process group and session leader to the own process.
669 * Add some fatal-error log message for failed malloc().
670 * Add '-d' if compiled with 'configure --with-debug' to
671 * enable debug output.
673 * Revision 1.52 2001/10/26 20:11:20 jongfoster
674 * Fixing type mismatch
676 * Revision 1.51 2001/10/26 17:38:28 oes
679 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
680 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
681 * threads to call select() simultaneously. So, it's time to do a real, live,
682 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
683 * (native). Both versions will work, but using __OS2__ offers multi-threading.
685 * Revision 1.49 2001/10/23 21:41:35 jongfoster
686 * Added call to initialize the (statically-allocated of course)
687 * "out of memory" CGI response.
689 * Revision 1.48 2001/10/10 19:56:46 jongfoster
690 * Moving some code that wasn't cookie-related out of an #ifdef
693 * Revision 1.47 2001/10/10 16:44:36 oes
694 * Added CONNECT destination port limitation check
696 * Revision 1.46 2001/10/08 15:17:41 oes
697 * Re-enabled SSL forwarding
699 * Revision 1.45 2001/10/07 15:42:11 oes
700 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
702 * Moved downgrading of the HTTP version from parse_http_request to
703 * chat(), since we can't decide if it is necessary before we have
704 * determined the actions for the URL. The HTTP command is now
705 * *always* re-built so the repairs need no longer be special-cased.
707 * filter_popups now gets a csp pointer so it can raise the new
708 * CSP_FLAG_MODIFIED flag.
712 * Added configurable size limit for the IOB. If the IOB grows so
713 * large that the next read would exceed the limit, the header
714 * is generated, and the header & unfiltered buffer are flushed
715 * to the client. Chat then continues in non-buffering,
716 * non-filtering body mode.
718 * Revision 1.44 2001/10/02 18:13:57 oes
721 * Revision 1.43 2001/10/02 15:32:13 oes
722 * Moved generation of hdr
724 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
725 * Cleaning up 2 compiler warnings on OS/2.
727 * Revision 1.41 2001/09/16 17:05:14 jongfoster
728 * Removing unused #include showarg.h
730 * Revision 1.40 2001/09/16 15:41:45 jongfoster
731 * Fixing signed/unsigned comparison warning.
733 * Revision 1.39 2001/09/16 13:21:27 jongfoster
734 * Changes to use new list functions.
736 * Revision 1.38 2001/09/16 13:01:46 jongfoster
737 * Removing redundant function call that zeroed zalloc()'d memory.
739 * Revision 1.37 2001/09/10 11:12:24 oes
740 * Deleted unused variable
742 * Revision 1.36 2001/09/10 10:56:15 oes
743 * Silenced compiler warnings
745 * Revision 1.35 2001/07/31 14:44:22 oes
746 * Deleted unused size parameter from filter_popups()
748 * Revision 1.34 2001/07/30 22:08:36 jongfoster
749 * Tidying up #defines:
750 * - All feature #defines are now of the form FEATURE_xxx
751 * - Permanently turned off WIN_GUI_EDIT
752 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
754 * Revision 1.33 2001/07/29 19:32:00 jongfoster
755 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
757 * Revision 1.32 2001/07/29 18:47:05 jongfoster
758 * Adding missing #include "loadcfg.h"
760 * Revision 1.31 2001/07/29 12:17:48 oes
761 * Applied pthread fix by Paul Lieverse
763 * Revision 1.30 2001/07/25 22:57:13 jongfoster
764 * __BEOS__ no longer overrides FEATURE_PTHREAD.
765 * This is because FEATURE_PTHREAD will soon be widely used, so I
766 * want to keep it simple.
768 * Revision 1.29 2001/07/24 12:47:06 oes
769 * Applied BeOS support update by Eugenia
771 * Revision 1.28 2001/07/23 13:26:12 oes
772 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
774 * Revision 1.27 2001/07/19 19:09:47 haroon
775 * - Added code to take care of the situation where while processing the first
776 * server response (which includes the server header), after finding the end
777 * of the headers we were not looking past the end of the headers for
778 * content modification. I enabled it for filter_popups.
779 * Someone else should look to see if other similar operations should be
780 * done to the discarded portion of the buffer.
782 * Note 2001/07/20: No, the other content modification mechanisms will process
783 * the whole iob later anyway. --oes
785 * Revision 1.26 2001/07/18 12:31:36 oes
788 * Revision 1.25 2001/07/15 19:43:49 jongfoster
789 * Supports POSIX threads.
790 * Also removed some unused #includes.
792 * Revision 1.24 2001/07/13 14:00:40 oes
793 * - Generic content modification scheme:
794 * Each feature has its own applicability flag that is set
795 * from csp->action->flags.
796 * Replaced the "filtering" int flag , by a function pointer
797 * "content_filter" to the function that will do the content
798 * modification. If it is != NULL, the document will be buffered
799 * and processed through *content_filter, which must set
800 * csp->content_length and return a modified copy of the body
801 * or return NULL (on failiure).
802 * - Changed csp->is_text to the more generic bitmap csp->content_type
803 * which can currently take the valued CT_TEXT or CT_GIF
805 * - Removed all #ifdef PCRS
807 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
808 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
809 * the `block_acl' call. This prevents a compilation error when the user
810 * does not wish to use the "ACL" feature.
812 * Revision 1.22 2001/06/29 21:45:41 oes
813 * Indentation, CRLF->LF, Tab-> Space
815 * Revision 1.21 2001/06/29 13:29:36 oes
816 * - Cleaned up, improved comments
817 * - Unified all possible interceptors (CGI,
818 * block, trust, fast_redirect) in one
819 * place, with one (CGI) answer generation
820 * mechansim. Much clearer now.
821 * - Removed the GIF image generation, which
822 * is now done in filters.c:block_url()
823 * - Made error conditions like domain lookup
824 * failiure or (various) problems while talking
825 * to the server use cgi.c:error_response()
826 * instead of generating HTML/HTTP in chat() (yuck!)
827 * - Removed logentry from cancelled commit
829 * Revision 1.20 2001/06/09 10:55:28 jongfoster
830 * Changing BUFSIZ ==> BUFFER_SIZE
832 * Revision 1.19 2001/06/07 23:12:52 jongfoster
833 * Replacing function pointer in struct gateway with a directly
834 * called function forwarded_connect().
835 * Replacing struct gateway with struct forward_spec
837 * Revision 1.18 2001/06/03 19:12:16 oes
838 * introduced new cgi handling
840 * Revision 1.17 2001/06/01 20:07:23 jongfoster
841 * Now uses action +image-blocker{} rather than config->tinygif
843 * Revision 1.16 2001/06/01 18:49:17 jongfoster
844 * Replaced "list_share" with "list" - the tiny memory gain was not
845 * worth the extra complexity.
847 * Revision 1.15 2001/05/31 21:24:47 jongfoster
848 * Changed "permission" to "action" throughout.
849 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
850 * Moved vanilla wafer check into chat(), since we must now
851 * decide whether or not to add it based on the URL.
853 * Revision 1.14 2001/05/29 20:14:01 joergs
854 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
857 * Revision 1.13 2001/05/29 09:50:24 jongfoster
858 * Unified blocklist/imagelist/permissionslist.
859 * File format is still under discussion, but the internal changes
862 * Also modified interceptor behaviour:
863 * - We now intercept all URLs beginning with one of the following
864 * prefixes (and *only* these prefixes):
866 * * http://ijbswa.sf.net/config/
867 * * http://ijbswa.sourceforge.net/config/
868 * - New interceptors "home page" - go to http://i.j.b/ to see it.
869 * - Internal changes so that intercepted and fast redirect pages
870 * are not replaced with an image.
871 * - Interceptors now have the option to send a binary page direct
872 * to the client. (i.e. ijb-send-banner uses this)
873 * - Implemented show-url-info interceptor. (Which is why I needed
874 * the above interceptors changes - a typical URL is
875 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
876 * The previous mechanism would not have intercepted that, and
877 * if it had been intercepted then it then it would have replaced
880 * Revision 1.12 2001/05/27 22:17:04 oes
882 * - re_process_buffer no longer writes the modified buffer
883 * to the client, which was very ugly. It now returns the
884 * buffer, which it is then written by chat.
886 * - content_length now adjusts the Content-Length: header
887 * for modified documents rather than crunch()ing it.
888 * (Length info in csp->content_length, which is 0 for
889 * unmodified documents)
891 * - For this to work, sed() is called twice when filtering.
893 * Revision 1.11 2001/05/26 17:27:53 jongfoster
894 * Added support for CLF and fixed LOG_LEVEL_LOG.
895 * Also did CRLF->LF fix of my previous patch.
897 * Revision 1.10 2001/05/26 15:26:15 jongfoster
898 * ACL feature now provides more security by immediately dropping
899 * connections from untrusted hosts.
901 * Revision 1.9 2001/05/26 00:28:36 jongfoster
902 * Automatic reloading of config file.
903 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
904 * Most of the global variables have been moved to a new
905 * struct configuration_spec, accessed through csp->config->globalname
906 * Most of the globals remaining are used by the Win32 GUI.
908 * Revision 1.8 2001/05/25 22:43:18 jongfoster
909 * Fixing minor memory leak and buffer overflow.
911 * Revision 1.7 2001/05/25 22:34:30 jongfoster
914 * Revision 1.6 2001/05/23 00:13:58 joergs
915 * AmigaOS support fixed.
917 * Revision 1.5 2001/05/22 18:46:04 oes
919 * - Enabled filtering banners by size rather than URL
920 * by adding patterns that replace all standard banner
921 * sizes with the "Junkbuster" gif to the re_filterfile
923 * - Enabled filtering WebBugs by providing a pattern
924 * which kills all 1x1 images
926 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
927 * which is selected by the (nonstandard and therefore
928 * capital) letter 'U' in the option string.
929 * It causes the quantifiers to be ungreedy by default.
930 * Appending a ? turns back to greedy (!).
932 * - Added a new interceptor ijb-send-banner, which
933 * sends back the "Junkbuster" gif. Without imagelist or
934 * MSIE detection support, or if tinygif = 1, or the
935 * URL isn't recognized as an imageurl, a lame HTML
936 * explanation is sent instead.
938 * - Added new feature, which permits blocking remote
939 * script redirects and firing back a local redirect
941 * The feature is conditionally compiled, i.e. it
942 * can be disabled with --disable-fast-redirects,
943 * plus it must be activated by a "fast-redirects"
944 * line in the config file, has its own log level
945 * and of course wants to be displayed by show-proxy-args
946 * Note: Boy, all the #ifdefs in 1001 locations and
947 * all the fumbling with configure.in and acconfig.h
948 * were *way* more work than the feature itself :-(
950 * - Because a generic redirect template was needed for
951 * this, tinygif = 3 now uses the same.
953 * - Moved GIFs, and other static HTTP response templates
958 * - Removed some >400 CRs again (Jon, you really worked
961 * Revision 1.4 2001/05/21 19:34:01 jongfoster
962 * Made failure to bind() a fatal error.
964 * Revision 1.3 2001/05/20 01:21:20 jongfoster
965 * Version 2.9.4 checkin.
966 * - Merged popupfile and cookiefile, and added control over PCRS
967 * filtering, in new "permissionsfile".
968 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
969 * file error you now get a message box (in the Win32 GUI) rather
970 * than the program exiting with no explanation.
971 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
973 * - Removed tabs from "config"
974 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
975 * - Bumped up version number.
977 * Revision 1.2 2001/05/17 22:34:44 oes
978 * - Added hint on GIF char array generation to jcc.c
979 * - Cleaned CRLF's from the sources and related files
980 * - Repaired logging for REF and FRC
982 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
983 * Initial import of version 2.9.3 source tree
986 *********************************************************************/
992 #include <sys/types.h>
1001 # ifndef FEATURE_PTHREAD
1005 # include <windows.h>
1006 # include <process.h>
1007 # endif /* ndef FEATURE_PTHREAD */
1010 # ifndef _WIN_CONSOLE
1011 # include "w32log.h"
1012 # endif /* ndef _WIN_CONSOLE */
1013 # include "w32svrapi.h"
1015 #else /* ifndef _WIN32 */
1017 # if !defined (__OS2__)
1018 # include <unistd.h>
1019 # include <sys/wait.h>
1020 # endif /* ndef __OS2__ */
1021 # include <sys/time.h>
1022 # include <sys/stat.h>
1023 # include <sys/ioctl.h>
1026 #include <sys/termios.h>
1034 # include <signal.h>
1037 # include <socket.h> /* BeOS has select() for sockets only. */
1038 # include <OS.h> /* declarations for threads and stuff. */
1041 # if defined(__EMX__) || defined(__OS2__)
1042 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1047 #define bzero(B,N) memset(B,0x00,n)
1051 # include <select.h>
1056 #include "project.h"
1059 #include "filters.h"
1060 #include "loaders.h"
1061 #include "parsers.h"
1062 #include "miscutil.h"
1064 #include "jbsockets.h"
1065 #include "gateway.h"
1066 #include "actions.h"
1068 #include "loadcfg.h"
1069 #include "urlmatch.h"
1071 const char jcc_h_rcs[] = JCC_H_VERSION;
1072 const char project_h_rcs[] = PROJECT_H_VERSION;
1075 struct client_state clients[1];
1076 struct file_list files[1];
1078 #ifdef FEATURE_STATISTICS
1079 int urls_read = 0; /* total nr of urls read inc rejected */
1080 int urls_rejected = 0; /* total nr of urls rejected */
1081 #endif /* def FEATURE_STATISTICS */
1083 #ifdef FEATURE_GRACEFUL_TERMINATION
1084 int g_terminate = 0;
1087 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1088 static void sig_handler(int the_signal);
1090 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1091 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1092 static jb_err get_server_headers(struct client_state *csp);
1093 static const char *crunch_reason(const struct http_response *rsp);
1094 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
1096 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1098 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1099 static jb_err change_request_destination(struct client_state *csp);
1100 static void chat(struct client_state *csp);
1101 static void serve(struct client_state *csp);
1102 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1103 static void usage(const char *myname);
1105 static void initialize_mutexes(void);
1106 static jb_socket bind_port_helper(struct configuration_spec *config);
1107 static void listen_loop(void);
1110 void serve(struct client_state *csp);
1111 #else /* ifndef AMIGA */
1112 static void serve(struct client_state *csp);
1113 #endif /* def AMIGA */
1116 static int32 server_thread(void *data);
1117 #endif /* def __BEOS__ */
1120 #define sleep(N) Sleep(((N) * 1000))
1124 #define sleep(N) DosSleep(((N) * 100))
1127 #ifdef FEATURE_PTHREAD
1128 pthread_mutex_t log_mutex;
1129 pthread_mutex_t log_init_mutex;
1131 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1132 pthread_mutex_t resolver_mutex;
1133 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1135 #ifndef HAVE_GMTIME_R
1136 pthread_mutex_t gmtime_mutex;
1137 #endif /* ndef HAVE_GMTIME_R */
1139 #ifndef HAVE_LOCALTIME_R
1140 pthread_mutex_t localtime_mutex;
1141 #endif /* ndef HAVE_GMTIME_R */
1144 pthread_mutex_t rand_mutex;
1145 #endif /* ndef HAVE_RANDOM */
1147 #endif /* FEATURE_PTHREAD */
1149 #if defined(unix) || defined(__EMX__)
1150 const char *basedir = NULL;
1151 const char *pidfile = NULL;
1152 int received_hup_signal = 0;
1153 #endif /* defined unix */
1155 /* HTTP snipplets. */
1156 static const char CSUCCEED[] =
1157 "HTTP/1.0 200 Connection established\r\n"
1158 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1160 static const char CHEADER[] =
1161 "HTTP/1.0 400 Invalid header received from client\r\n"
1162 "Proxy-Agent: Privoxy " VERSION "\r\n"
1163 "Content-Type: text/plain\r\n"
1164 "Connection: close\r\n\r\n"
1165 "Invalid header received from client.\r\n";
1167 static const char FTP_RESPONSE[] =
1168 "HTTP/1.0 400 Invalid request received from client\r\n"
1169 "Content-Type: text/plain\r\n"
1170 "Connection: close\r\n\r\n"
1171 "Invalid request. Privoxy doesn't support FTP.\r\n";
1173 static const char GOPHER_RESPONSE[] =
1174 "HTTP/1.0 400 Invalid request received from client\r\n"
1175 "Content-Type: text/plain\r\n"
1176 "Connection: close\r\n\r\n"
1177 "Invalid request. Privoxy doesn't support gopher.\r\n";
1179 /* XXX: should be a template */
1180 static const char MISSING_DESTINATION_RESPONSE[] =
1181 "HTTP/1.0 400 Bad request received from client\r\n"
1182 "Proxy-Agent: Privoxy " VERSION "\r\n"
1183 "Content-Type: text/plain\r\n"
1184 "Connection: close\r\n\r\n"
1185 "Bad request. Privoxy was unable to extract the destination.\r\n";
1187 /* XXX: should be a template */
1188 static const char NO_SERVER_DATA_RESPONSE[] =
1189 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1190 "Proxy-Agent: Privoxy " VERSION "\r\n"
1191 "Content-Type: text/plain\r\n"
1192 "Connection: close\r\n\r\n"
1193 "Empty server or forwarder response.\r\n"
1194 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1196 /* XXX: should be a template */
1197 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1198 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1199 "Proxy-Agent: Privoxy " VERSION "\r\n"
1200 "Content-Type: text/plain\r\n"
1201 "Connection: close\r\n\r\n"
1202 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1205 /* XXX: should be a template */
1206 static const char NULL_BYTE_RESPONSE[] =
1207 "HTTP/1.0 400 Bad request received from client\r\n"
1208 "Proxy-Agent: Privoxy " VERSION "\r\n"
1209 "Content-Type: text/plain\r\n"
1210 "Connection: close\r\n\r\n"
1211 "Bad request. Null byte(s) before end of request.\r\n";
1214 /* XXX: should be a template */
1215 static const char MESSED_UP_REQUEST_RESPONSE[] =
1216 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1217 "Proxy-Agent: Privoxy " VERSION "\r\n"
1218 "Content-Type: text/plain\r\n"
1219 "Connection: close\r\n\r\n"
1220 "Bad request. Messed up with header filters.\r\n";
1222 /* A function to crunch a response */
1223 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1225 /* Crunch function flags */
1226 #define CF_NO_FLAGS 0
1227 /* Cruncher applies to forced requests as well */
1228 #define CF_IGNORE_FORCE 1
1229 /* Crunched requests are counted for the block statistics */
1230 #define CF_COUNT_AS_REJECT 2
1232 /* A crunch function and its flags */
1235 const crunch_func_ptr cruncher;
1239 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1241 /* Complete list of cruncher functions */
1242 static const struct cruncher crunchers_all[] = {
1243 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1244 { block_url, CF_COUNT_AS_REJECT },
1245 #ifdef FEATURE_TRUST
1246 { trust_url, CF_COUNT_AS_REJECT },
1247 #endif /* def FEATURE_TRUST */
1248 { redirect_url, CF_NO_FLAGS },
1249 { dispatch_cgi, CF_IGNORE_FORCE},
1253 /* Light version, used after tags are applied */
1254 static const struct cruncher crunchers_light[] = {
1255 { block_url, CF_COUNT_AS_REJECT },
1256 { redirect_url, CF_NO_FLAGS },
1261 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1262 /*********************************************************************
1264 * Function : sig_handler
1266 * Description : Signal handler for different signals.
1267 * Exit gracefully on TERM and INT
1268 * or set a flag that will cause the errlog
1269 * to be reopened by the main thread on HUP.
1272 * 1 : the_signal = the signal cause this function to call
1276 *********************************************************************/
1277 static void sig_handler(int the_signal)
1283 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1294 received_hup_signal = 1;
1299 * We shouldn't be here, unless we catch signals
1300 * in main() that we can't handle here!
1302 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1310 /*********************************************************************
1312 * Function : client_protocol_is_unsupported
1314 * Description : Checks if the client used a known unsupported
1315 * protocol and deals with it by sending an error
1319 * 1 : csp = Current client state (buffers, headers, etc...)
1320 * 2 : req = the first request line send by the client
1322 * Returns : TRUE if an error response has been generated, or
1323 * FALSE if the request doesn't look invalid.
1325 *********************************************************************/
1326 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1329 * If it's a FTP or gopher request, we don't support it.
1331 * These checks are better than nothing, but they might
1332 * not work in all configurations and some clients might
1333 * have problems digesting the answer.
1335 * They should, however, never cause more problems than
1336 * Privoxy's old behaviour (returning the misleading HTML
1339 * "Could not resolve http://(ftp|gopher)://example.org").
1341 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1343 const char *response = NULL;
1344 const char *protocol = NULL;
1346 if (!strncmpic(req, "GET ftp://", 10))
1348 response = FTP_RESPONSE;
1353 response = GOPHER_RESPONSE;
1354 protocol = "GOPHER";
1356 log_error(LOG_LEVEL_ERROR,
1357 "%s tried to use Privoxy as %s proxy: %s",
1358 csp->ip_addr_str, protocol, req);
1359 log_error(LOG_LEVEL_CLF,
1360 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1362 write_socket(csp->cfd, response, strlen(response));
1371 /*********************************************************************
1373 * Function : get_request_destination_elsewhere
1375 * Description : If the client's request was redirected into
1376 * Privoxy without the client's knowledge,
1377 * the request line lacks the destination host.
1379 * This function tries to get it elsewhere,
1380 * provided accept-intercepted-requests is enabled.
1382 * "Elsewhere" currently only means "Host: header",
1383 * but in the future we may ask the redirecting
1384 * packet filter to look the destination up.
1386 * If the destination stays unknown, an error
1387 * response is send to the client and headers
1388 * are freed so that chat() can return directly.
1391 * 1 : csp = Current client state (buffers, headers, etc...)
1392 * 2 : headers = a header list
1394 * Returns : JB_ERR_OK if the destination is now known, or
1395 * JB_ERR_PARSE if it isn't.
1397 *********************************************************************/
1398 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1402 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1404 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1405 " Privoxy isn't configured to accept intercepted requests.",
1406 csp->ip_addr_str, csp->http->cmd);
1407 /* XXX: Use correct size */
1408 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1409 csp->ip_addr_str, csp->http->cmd);
1411 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1412 destroy_list(headers);
1414 return JB_ERR_PARSE;
1416 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1418 /* Split the domain we just got for pattern matching */
1419 init_domain_components(csp->http);
1425 /* We can't work without destination. Go spread the news.*/
1427 req = list_to_text(headers);
1429 /* XXX: Use correct size */
1430 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1431 csp->ip_addr_str, csp->http->cmd);
1432 log_error(LOG_LEVEL_ERROR,
1433 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1434 csp->ip_addr_str, csp->http->cmd, req);
1437 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1438 destroy_list(headers);
1440 return JB_ERR_PARSE;
1443 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1444 * to get the destination IP address, use it as host directly
1445 * or do a reverse DNS lookup first.
1450 /*********************************************************************
1452 * Function : get_server_headers
1454 * Description : Parses server headers in iob and fills them
1455 * into csp->headers so that they can later be
1459 * 1 : csp = Current client state (buffers, headers, etc...)
1461 * Returns : JB_ERR_OK if everything went fine, or
1462 * JB_ERR_PARSE if the headers were incomplete.
1464 *********************************************************************/
1465 static jb_err get_server_headers(struct client_state *csp)
1467 int continue_hack_in_da_house = 0;
1470 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1475 * continue hack in da house. Ignore the ending of
1476 * this head and continue enlisting header lines.
1477 * The reason is described below.
1479 enlist(csp->headers, "");
1480 continue_hack_in_da_house = 0;
1483 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1486 * It's a bodyless continue response, don't
1487 * stop header parsing after reaching it's end.
1489 * As a result Privoxy will concatenate the
1490 * next response's head and parse and deliver
1491 * the headers as if they belonged to one request.
1493 * The client will separate them because of the
1494 * empty line between them.
1496 * XXX: What we're doing here is clearly against
1497 * the intended purpose of the continue header,
1498 * and under some conditions (HTTP/1.0 client request)
1499 * it's a standard violation.
1501 * Anyway, "sort of against the spec" is preferable
1502 * to "always getting confused by Continue responses"
1503 * (Privoxy's behaviour before this hack was added)
1505 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1506 continue_hack_in_da_house = 1;
1508 else if (*header == '\0')
1511 * If the header is empty, but the Continue hack
1512 * isn't active, we can assume that we reached the
1513 * end of the buffer before we hit the end of the
1516 * Inform the caller an let it decide how to handle it.
1518 return JB_ERR_PARSE;
1522 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1525 * XXX: Should we quit the request and return a
1526 * out of memory error page instead?
1528 log_error(LOG_LEVEL_ERROR,
1529 "Out of memory while enlisting server headers. %s lost.",
1539 /*********************************************************************
1541 * Function : crunch_reason
1543 * Description : Translates the crunch reason code into a string.
1546 * 1 : rsp = a http_response
1548 * Returns : A string with the crunch reason or an error description.
1550 *********************************************************************/
1551 static const char *crunch_reason(const struct http_response *rsp)
1553 char * reason = NULL;
1555 assert(rsp != NULL);
1558 return "Internal error while searching for crunch reason";
1561 switch (rsp->reason)
1563 case RSP_REASON_UNSUPPORTED:
1564 reason = "Unsupported HTTP feature";
1566 case RSP_REASON_BLOCKED:
1569 case RSP_REASON_UNTRUSTED:
1570 reason = "Untrusted";
1572 case RSP_REASON_REDIRECTED:
1573 reason = "Redirected";
1575 case RSP_REASON_CGI_CALL:
1576 reason = "CGI Call";
1578 case RSP_REASON_NO_SUCH_DOMAIN:
1579 reason = "DNS failure";
1581 case RSP_REASON_FORWARDING_FAILED:
1582 reason = "Forwarding failed";
1584 case RSP_REASON_CONNECT_FAILED:
1585 reason = "Connection failure";
1587 case RSP_REASON_OUT_OF_MEMORY:
1588 reason = "Out of memory (may mask other reasons)";
1591 reason = "No reason recorded";
1599 /*********************************************************************
1601 * Function : send_crunch_response
1603 * Description : Delivers already prepared response for
1604 * intercepted requests, logs the interception
1605 * and frees the response.
1608 * 1 : csp = Current client state (buffers, headers, etc...)
1609 * 1 : rsp = Fully prepared response. Will be freed on exit.
1611 * Returns : Nothing.
1613 *********************************************************************/
1614 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1616 const struct http_request *http = csp->http;
1617 char status_code[4];
1619 assert(rsp != NULL);
1620 assert(rsp->head != NULL);
1625 * Not supposed to happen. If it does
1626 * anyway, treat it as an unknown error.
1628 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1629 /* return code doesn't matter */
1634 /* If rsp is still NULL, we have serious internal problems. */
1635 log_error(LOG_LEVEL_FATAL,
1636 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1640 * Extract the status code from the actual head
1641 * that was send to the client. It is the only
1642 * way to get it right for all requests, including
1643 * the fixed ones for out-of-memory problems.
1645 * A head starts like this: 'HTTP/1.1 200...'
1649 status_code[0] = rsp->head[9];
1650 status_code[1] = rsp->head[10];
1651 status_code[2] = rsp->head[11];
1652 status_code[3] = '\0';
1654 /* Write the answer to the client */
1655 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1656 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1658 /* There is nothing we can do about it. */
1659 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1662 /* Log that the request was crunched and why. */
1663 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1664 http->hostport, http->path, crunch_reason(rsp));
1665 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1666 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1668 /* Clean up and return */
1669 if (cgi_error_memory() != rsp)
1671 free_http_response(rsp);
1678 /*********************************************************************
1680 * Function : request_contains_null_bytes
1682 * Description : Checks for NULL bytes in the request and sends
1683 * an error message to the client if any were found.
1685 * XXX: currently not used, see comment in chat().
1688 * 1 : csp = Current client state (buffers, headers, etc...)
1689 * 2 : buf = Data from the client's request to check.
1690 * 3 : len = The data length.
1692 * Returns : TRUE if the request contained one or more NULL bytes, or
1695 *********************************************************************/
1696 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1698 size_t c_len; /* Request lenght when treated as C string */
1700 c_len = strlen(buf);
1705 * Null byte(s) found. Log the request,
1706 * return an error response and hang up.
1708 size_t tmp_len = c_len;
1713 * Replace NULL byte(s) with '°' characters
1714 * so the request can be logged as string.
1715 * XXX: Is there a better replacement character?
1718 tmp_len += strlen(buf+tmp_len);
1719 } while (tmp_len < len);
1721 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1722 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1723 log_error(LOG_LEVEL_HEADER,
1724 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1726 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1728 /* XXX: Log correct size */
1729 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1739 /*********************************************************************
1741 * Function : crunch_response_triggered
1743 * Description : Checks if the request has to be crunched,
1744 * and delivers the crunch response if necessary.
1747 * 1 : csp = Current client state (buffers, headers, etc...)
1748 * 2 : crunchers = list of cruncher functions to run
1750 * Returns : TRUE if the request was answered with a crunch response
1753 *********************************************************************/
1754 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1756 struct http_response *rsp = NULL;
1757 const struct cruncher *c;
1760 * If CGI request crunching is disabled,
1761 * check the CGI dispatcher out of order to
1762 * prevent unintentional blocks or redirects.
1764 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1765 && (NULL != (rsp = dispatch_cgi(csp))))
1767 /* Deliver, log and free the interception response. */
1768 send_crunch_response(csp, rsp);
1772 for (c = crunchers; c->cruncher != NULL; c++)
1775 * Check the cruncher if either Privoxy is toggled
1776 * on and the request isn't forced, or if the cruncher
1777 * applies to forced requests as well.
1779 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1780 !(csp->flags & CSP_FLAG_FORCED)) ||
1781 (c->flags & CF_IGNORE_FORCE))
1783 rsp = c->cruncher(csp);
1786 /* Deliver, log and free the interception response. */
1787 send_crunch_response(csp, rsp);
1788 #ifdef FEATURE_STATISTICS
1789 if (c->flags & CF_COUNT_AS_REJECT)
1791 csp->flags |= CSP_FLAG_REJECTED;
1793 #endif /* def FEATURE_STATISTICS */
1804 /*********************************************************************
1806 * Function : build_request_line
1808 * Description : Builds the HTTP request line.
1810 * If a HTTP forwarder is used it expects the whole URL,
1811 * web servers only get the path.
1814 * 1 : csp = Current client state (buffers, headers, etc...)
1815 * 2 : fwd = The forwarding spec used for the request
1816 * XXX: Should use http->fwd instead.
1817 * 3 : request_line = The old request line which will be replaced.
1819 * Returns : Nothing. Terminates in case of memory problems.
1821 *********************************************************************/
1822 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1824 struct http_request *http = csp->http;
1826 assert(http->ssl == 0);
1829 * Downgrade http version from 1.1 to 1.0
1830 * if +downgrade action applies.
1832 if ( (csp->action->flags & ACTION_DOWNGRADE)
1833 && (!strcmpic(http->ver, "HTTP/1.1")))
1836 http->ver = strdup("HTTP/1.0");
1838 if (http->ver == NULL)
1840 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1845 * Rebuild the request line.
1847 freez(*request_line);
1848 *request_line = strdup(http->gpc);
1849 string_append(request_line, " ");
1851 if (fwd->forward_host)
1853 string_append(request_line, http->url);
1857 string_append(request_line, http->path);
1859 string_append(request_line, " ");
1860 string_append(request_line, http->ver);
1862 if (*request_line == NULL)
1864 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1866 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1870 /*********************************************************************
1872 * Function : change_request_destination
1874 * Description : Parse a (rewritten) request line and regenerate
1875 * the http request data.
1878 * 1 : csp = Current client state (buffers, headers, etc...)
1880 * Returns : Forwards the parse_http_request() return code.
1881 * Terminates in case of memory problems.
1883 *********************************************************************/
1884 static jb_err change_request_destination(struct client_state *csp)
1886 struct http_request *http = csp->http;
1889 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1890 free_http_request(http);
1891 err = parse_http_request(csp->headers->first->str, http, csp);
1892 if (JB_ERR_OK != err)
1894 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1895 jb_err_to_string(err));
1897 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1898 if (http->ocmd == NULL)
1900 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1907 /*********************************************************************
1911 * Description : Once a connection to the client has been accepted,
1912 * this function is called (via serve()) to handle the
1913 * main business of the communication. When this
1914 * function returns, the caller must close the client
1917 * FIXME: chat is nearly thousand lines long.
1921 * 1 : csp = Current client state (buffers, headers, etc...)
1923 * Returns : Nothing.
1925 *********************************************************************/
1926 static void chat(struct client_state *csp)
1928 char buf[BUFFER_SIZE];
1936 int ms_iis5_hack = 0;
1937 size_t byte_count = 0;
1938 int forwarded_connect_retries = 0;
1939 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1940 const struct forward_spec * fwd;
1941 struct http_request *http;
1942 int len; /* for buffer sizes (and negative error codes) */
1945 /* Function that does the content filtering for the current request */
1946 filter_function_ptr content_filter = NULL;
1948 /* Skeleton for HTTP response, if we should intercept the request */
1949 struct http_response *rsp;
1951 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1952 struct list header_list;
1953 struct list *headers = &header_list;
1957 memset(buf, 0, sizeof(buf));
1960 * Read the client's request. Note that since we're not using select() we
1961 * could get blocked here if a client connected, then didn't say anything!
1966 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1968 if (len <= 0) break; /* error! */
1971 * If there is no memory left for buffering the
1972 * request, there is nothing we can do but hang up
1974 if (add_to_iob(csp, buf, len))
1979 req = get_header(csp);
1981 } while ((NULL != req) && ('\0' == *req));
1983 if ((NULL != req) && ('\0' != *req))
1985 /* Request received. Validate and parse it. */
1989 * XXX: Temporary disabled to prevent problems
1990 * with POST requests whose bodies are allowed to
1991 * contain NULL bytes. BR#1730105.
1993 * The main purpose of this check is to properly
1994 * log stuff like BitTorrent traffic and other junk
1995 * that hits public proxies. It's not required for
1996 * Privoxy to functions as those requests are discarded
1999 * It probably should be rewritten to only check
2000 * the head of the request. Another option would
2001 * be to let all POST requests pass, although that
2002 * may not be good enough.
2004 if (request_contains_null_bytes(csp, buf, len))
2006 /* NULL bytes found and dealt with, just hang up. */
2011 /* Does the request line look invalid? */
2012 if (client_protocol_is_unsupported(csp, req))
2015 * Yes. The request has already been
2016 * answered with a error response, the buffers
2017 * were freed and we're done with chatting.
2022 #ifdef FEATURE_FORCE_LOAD
2024 * If this request contains the FORCE_PREFIX and blocks
2025 * aren't enforced, get rid of it and set the force flag.
2027 if (strstr(req, FORCE_PREFIX))
2029 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2031 log_error(LOG_LEVEL_FORCE,
2032 "Ignored force prefix in request: \"%s\".", req);
2036 strclean(req, FORCE_PREFIX);
2037 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2038 csp->flags |= CSP_FLAG_FORCED;
2042 #endif /* def FEATURE_FORCE_LOAD */
2043 err = parse_http_request(req, http, csp);
2044 if (JB_ERR_OK != err)
2046 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2052 if (http->cmd == NULL)
2054 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2055 /* XXX: Use correct size */
2056 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2057 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2059 free_http_request(http);
2063 /* grab the rest of the client's headers */
2067 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
2069 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2072 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2073 destroy_list(headers);
2078 * If there is no memory left for buffering the
2079 * request, there is nothing we can do but hang up
2081 if (add_to_iob(csp, buf, len))
2083 destroy_list(headers);
2089 if (p == NULL) break;
2096 if (http->host == NULL)
2099 * If we still don't know the request destination,
2100 * the request is invalid or the client uses
2101 * Privoxy without its knowledge.
2103 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2106 * Our attempts to get the request destination
2107 * elsewhere failed or Privoxy is configured
2108 * to only accept proxy requests.
2110 * An error response has already been send
2111 * and we're done here.
2118 * Determine the actions for this URL
2120 #ifdef FEATURE_TOGGLE
2121 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2123 /* Most compatible set of actions (i.e. none) */
2124 init_current_action(csp->action);
2127 #endif /* ndef FEATURE_TOGGLE */
2129 get_url_actions(csp, http);
2133 * Save a copy of the original request for logging
2135 http->ocmd = strdup(http->cmd);
2137 if (http->ocmd == NULL)
2139 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2142 enlist(csp->headers, http->cmd);
2144 /* Append the previously read headers */
2145 list_append_list_unique(csp->headers, headers);
2146 destroy_list(headers);
2148 err = sed(client_patterns, add_client_headers, csp);
2149 if (JB_ERR_OK != err)
2151 assert(err == JB_ERR_PARSE);
2152 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2154 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2156 /* Check request line for rewrites. */
2157 if ((NULL == csp->headers->first->str)
2158 || (strcmp(http->cmd, csp->headers->first->str) &&
2159 (JB_ERR_OK != change_request_destination(csp))))
2162 * A header filter broke the request line - bail out.
2164 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2165 /* XXX: Use correct size */
2166 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2167 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2169 free_http_request(http);
2173 /* decide how to route the HTTP request */
2174 fwd = forward_url(csp, http);
2177 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2178 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2182 * build the http request to send to the server
2183 * we have to do one of the following:
2185 * create = use the original HTTP request to create a new
2186 * HTTP request that has either the path component
2187 * without the http://domainspec (w/path) or the
2188 * full orininal URL (w/url)
2189 * Note that the path and/or the HTTP version may
2190 * have been altered by now.
2192 * connect = Open a socket to the host:port of the server
2193 * and short-circuit server and client socket.
2195 * pass = Pass the request unchanged if forwarding a CONNECT
2196 * request to a parent proxy. Note that we'll be sending
2197 * the CFAIL message ourselves if connecting to the parent
2198 * fails, but we won't send a CSUCCEED message if it works,
2199 * since that would result in a double message (ours and the
2200 * parent's). After sending the request to the parent, we simply
2203 * here's the matrix:
2206 * +--------+--------+
2208 * 0 | create | connect|
2210 * Forwarding +--------+--------+
2212 * 1 | create | pass |
2214 * +--------+--------+
2218 if (http->ssl && connect_port_is_forbidden(csp))
2220 const char *acceptable_connect_ports =
2221 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2222 assert(NULL != acceptable_connect_ports);
2223 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2224 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2225 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2226 csp->action->flags |= ACTION_BLOCK;
2232 freez(csp->headers->first->str);
2233 build_request_line(csp, fwd, &csp->headers->first->str);
2236 hdr = list_to_text(csp->headers);
2239 /* FIXME Should handle error properly */
2240 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2244 * We have a request. Check if one of the crunchers wants it.
2246 if (crunch_response_triggered(csp, crunchers_all))
2249 * Yes. The client got the crunch response
2250 * and we are done here after cleaning up.
2253 list_remove_all(csp->headers);
2259 * The headers can't be removed earlier because
2260 * they were still needed for the referrer check
2261 * in case of CGI crunches.
2263 * XXX: Would it be worth to move the referrer check
2264 * into client_referrer() and set a flag if it's trusted?
2266 list_remove_all(csp->headers);
2268 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2270 if (fwd->forward_host)
2272 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2273 fwd->forward_host, fwd->forward_port, http->hostport);
2277 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2280 /* here we connect to the server, gateway, or the forwarder */
2282 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2283 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2285 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2286 forwarded_connect_retries, http->hostport);
2289 if (csp->sfd == JB_INVALID_SOCKET)
2291 if (fwd->type != SOCKS_NONE)
2294 rsp = error_response(csp, "forwarding-failed", errno);
2296 else if (errno == EINVAL)
2298 rsp = error_response(csp, "no-such-domain", errno);
2302 rsp = error_response(csp, "connect-failed", errno);
2303 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2308 /* Write the answer to the client */
2311 send_crunch_response(csp, rsp);
2318 if (fwd->forward_host || (http->ssl == 0))
2320 /* write the client's (modified) header to the server
2321 * (along with anything else that may be in the buffer)
2324 if (write_socket(csp->sfd, hdr, strlen(hdr))
2325 || (flush_socket(csp->sfd, csp->iob) < 0))
2327 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2330 rsp = error_response(csp, "connect-failed", errno);
2334 send_crunch_response(csp, rsp);
2344 * We're running an SSL tunnel and we're not forwarding,
2345 * so just send the "connect succeeded" message to the
2346 * client, flush the rest, and get out of the way.
2348 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2356 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2358 /* we're finished with the client's header */
2361 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2363 /* pass data between the client and server
2364 * until one or the other shuts down the connection.
2373 * FD_ZERO here seems to point to an errant macro which crashes.
2374 * So do this by hand for now...
2376 memset(&rfds,0x00,sizeof(fd_set));
2380 FD_SET(csp->cfd, &rfds);
2381 FD_SET(csp->sfd, &rfds);
2383 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2387 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2391 /* this is the body of the browser's request
2392 * just read it and write it.
2395 if (FD_ISSET(csp->cfd, &rfds))
2397 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2401 break; /* "game over, man" */
2404 if (write_socket(csp->sfd, buf, (size_t)len))
2406 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2413 * The server wants to talk. It could be the header or the body.
2414 * If `hdr' is null, then it's the header otherwise it's the body.
2415 * FIXME: Does `hdr' really mean `host'? No.
2419 if (FD_ISSET(csp->sfd, &rfds))
2422 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2426 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2428 if (http->ssl && (fwd->forward_host == NULL))
2431 * Just hang up. We already confirmed the client's CONNECT
2432 * request with status code 200 and unencrypted content is
2433 * no longer welcome.
2435 log_error(LOG_LEVEL_ERROR,
2436 "CONNECT already confirmed. Unable to tell the client about the problem.");
2439 else if (byte_count)
2442 * Just hang up. We already transmitted the original headers
2443 * and parts of the original content and therefore missed the
2444 * chance to send an error message (without risking data corruption).
2446 * XXX: we could retry with a fancy range request here.
2448 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2449 "Unable to tell the client about the problem.");
2453 rsp = error_response(csp, "connect-failed", errno);
2457 send_crunch_response(csp, rsp);
2463 /* Add a trailing zero. This lets filter_popups
2464 * use string operations.
2468 /* Normally, this would indicate that we've read
2469 * as much as the server has sent us and we can
2470 * close the client connection. However, Microsoft
2471 * in its wisdom has released IIS/5 with a bug that
2472 * prevents it from sending the trailing \r\n in
2473 * a 302 redirect header (and possibly other headers).
2474 * To work around this if we've haven't parsed
2475 * a full header we'll append a trailing \r\n
2476 * and see if this now generates a valid one.
2478 * This hack shouldn't have any impacts. If we've
2479 * already transmitted the header or if this is a
2480 * SSL connection, then we won't bother with this
2481 * hack. So we only work on partially received
2482 * headers. If we append a \r\n and this still
2483 * doesn't generate a valid header, then we won't
2484 * transmit anything to the client.
2489 if (server_body || http->ssl)
2492 * If we have been buffering up the document,
2493 * now is the time to apply content modification
2494 * and send the result to the client.
2498 p = execute_content_filter(csp, content_filter);
2500 * If the content filter fails, use the original
2501 * buffer and length.
2502 * (see p != NULL ? p : csp->iob->cur below)
2506 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2509 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2511 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2514 hdr = list_to_text(csp->headers);
2517 /* FIXME Should handle error properly */
2518 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2522 * Shouldn't happen because this was the second sed run
2523 * and tags are only created for the first one.
2525 assert(!crunch_response_triggered(csp, crunchers_all));
2527 if (write_socket(csp->cfd, hdr, strlen(hdr))
2528 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2530 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2540 break; /* "game over, man" */
2544 * This is NOT the body, so
2545 * Let's pretend the server just sent us a blank line.
2547 log_error(LOG_LEVEL_INFO,
2548 "Malformerd HTTP headers detected and MS IIS5 hack enabled. "
2549 "Expect an invalid response or even no response at all.");
2550 snprintf(buf, sizeof(buf), "\r\n");
2551 len = (int)strlen(buf);
2554 * Now, let the normal header parsing algorithm below do its
2555 * job. If it fails, we'll exit instead of continuing.
2562 * If this is an SSL connection or we're in the body
2563 * of the server document, just write it to the client,
2564 * unless we need to buffer the body for later content-filtering
2567 if (server_body || http->ssl)
2572 * If there is no memory left for buffering the content, or the buffer limit
2573 * has been reached, switch to non-filtering mode, i.e. make & write the
2574 * header, flush the iob and buf, and get out of the way.
2576 if (add_to_iob(csp, buf, len))
2581 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2583 hdr = list_to_text(csp->headers);
2587 * Memory is too tight to even generate the header.
2588 * Send our static "Out-of-memory" page.
2590 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2591 rsp = cgi_error_memory();
2592 send_crunch_response(csp, rsp);
2596 hdrlen = strlen(hdr);
2598 if (write_socket(csp->cfd, hdr, hdrlen)
2599 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2600 || (write_socket(csp->cfd, buf, (size_t)len)))
2602 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2609 * Reset the byte_count to the amount of bytes
2610 * we just flushed. len will be added a few lines below,
2611 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2613 byte_count = (size_t)flushed;
2615 content_filter = NULL;
2621 if (write_socket(csp->cfd, buf, (size_t)len))
2623 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2627 byte_count += (size_t)len;
2632 /* we're still looking for the end of the
2633 * server's header ... (does that make header
2634 * parsing an "out of body experience" ?
2638 * buffer up the data we just read. If that fails,
2639 * there's little we can do but send our static
2640 * out-of-memory page.
2642 if (add_to_iob(csp, buf, len))
2644 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2645 rsp = cgi_error_memory();
2646 send_crunch_response(csp, rsp);
2651 /* Convert iob into something sed() can digest */
2652 if (JB_ERR_PARSE == get_server_headers(csp))
2656 /* Well, we tried our MS IIS/5
2657 * hack and it didn't work.
2658 * The header is incomplete
2659 * and there isn't anything
2660 * we can do about it.
2666 /* Since we have to wait for
2667 * more from the server before
2668 * we can parse the headers
2669 * we just continue here.
2675 /* Did we actually get anything? */
2676 if (NULL == csp->headers->first)
2678 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2679 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2680 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2681 free_http_request(http);
2685 assert(csp->headers->first->str);
2687 if (strncmpic(csp->headers->first->str, "HTTP", 4))
2690 * It doesn't look like a HTTP response:
2691 * tell the client and log the problem.
2693 if (strlen(csp->headers->first->str) > 30)
2695 csp->headers->first->str[30] = '\0';
2697 log_error(LOG_LEVEL_ERROR,
2698 "Invalid server or forwarder response. Starts with: %s",
2699 csp->headers->first->str);
2700 log_error(LOG_LEVEL_CLF,
2701 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2702 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2703 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2704 free_http_request(http);
2708 /* we have now received the entire header.
2709 * filter it and send the result to the client
2711 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2713 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2715 hdr = list_to_text(csp->headers);
2718 /* FIXME Should handle error properly */
2719 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2722 if (crunch_response_triggered(csp, crunchers_light))
2725 * One of the tags created by a server-header
2726 * tagger triggered a crunch. We already
2727 * delivered the crunch response to the client
2728 * and are done here after cleaning up.
2733 /* Buffer and pcrs filter this if appropriate. */
2735 if (!http->ssl) /* We talk plaintext */
2737 content_filter = get_filter_function(csp);
2740 * Only write if we're not buffering for content modification
2742 if (!content_filter)
2744 /* write the server's (modified) header to
2745 * the client (along with anything else that
2746 * may be in the buffer)
2749 if (write_socket(csp->cfd, hdr, strlen(hdr))
2750 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2752 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2754 /* the write failed, so don't bother
2755 * mentioning it to the client...
2756 * it probably can't hear us anyway.
2762 byte_count += (size_t)len;
2765 /* we're finished with the server's header */
2770 /* If this was a MS IIS/5 hack then it means
2771 * the server has already closed the
2772 * connection. Nothing more to read. Time
2783 return; /* huh? we should never get here */
2786 if (csp->content_length == 0)
2789 * If Privoxy didn't recalculate the
2790 * Content-Lenght, byte_count is still
2793 csp->content_length = byte_count;
2796 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2797 csp->ip_addr_str, http->ocmd, csp->content_length);
2801 /*********************************************************************
2805 * Description : This is little more than chat. We only "serve" to
2806 * to close any socket that chat may have opened.
2809 * 1 : csp = Current client state (buffers, headers, etc...)
2813 *********************************************************************/
2815 void serve(struct client_state *csp)
2816 #else /* ifndef AMIGA */
2817 static void serve(struct client_state *csp)
2818 #endif /* def AMIGA */
2821 close_socket(csp->cfd);
2823 if (csp->sfd != JB_INVALID_SOCKET)
2825 close_socket(csp->sfd);
2828 csp->flags &= ~CSP_FLAG_ACTIVE;
2834 /*********************************************************************
2836 * Function : server_thread
2838 * Description : We only exist to call `serve' in a threaded environment.
2841 * 1 : data = Current client state (buffers, headers, etc...)
2843 * Returns : Always 0.
2845 *********************************************************************/
2846 static int32 server_thread(void *data)
2848 serve((struct client_state *) data);
2855 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2856 /*********************************************************************
2860 * Description : Print usage info & exit.
2862 * Parameters : Pointer to argv[0] for identifying ourselves
2866 *********************************************************************/
2867 static void usage(const char *myname)
2869 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2873 #endif /* defined(unix) */
2876 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2877 #endif /* defined(unix) */
2878 "[--version] [configfile]\n"
2879 "Aborting\n", myname);
2884 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2887 /*********************************************************************
2889 * Function : initialize_mutexes
2891 * Description : Prepares mutexes if mutex support is available.
2895 * Returns : Void, exits in case of errors.
2897 *********************************************************************/
2898 static void initialize_mutexes(void)
2902 #ifdef FEATURE_PTHREAD
2904 * Prepare global mutex semaphores
2906 err = pthread_mutex_init(&log_mutex, 0);
2908 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2911 * XXX: The assumptions below are a bit naive
2912 * and can cause locks that aren't necessary.
2914 * For example older FreeBSD versions (< 6.x?)
2915 * have no gethostbyname_r, but gethostbyname is
2918 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2919 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2920 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2922 * XXX: should we use a single mutex for
2923 * localtime() and gmtime() as well?
2925 #ifndef HAVE_GMTIME_R
2926 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2927 #endif /* ndef HAVE_GMTIME_R */
2929 #ifndef HAVE_LOCALTIME_R
2930 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2931 #endif /* ndef HAVE_GMTIME_R */
2934 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2935 #endif /* ndef HAVE_RANDOM */
2936 #endif /* FEATURE_PTHREAD */
2939 * TODO: mutex support for mingw32 would be swell.
2944 printf("Fatal error. Mutex initialization failed: %s.\n",
2953 /*********************************************************************
2957 * Description : Load the config file and start the listen loop.
2958 * This function is a lot more *sane* with the `load_config'
2959 * and `listen_loop' functions; although it stills does
2960 * a *little* too much for my taste.
2963 * 1 : argc = Number of parameters (including $0).
2964 * 2 : argv = Array of (char *)'s to the parameters.
2966 * Returns : 1 if : can't open config file, unrecognized directive,
2967 * stats requested in multi-thread mode, can't open the
2968 * log file, can't open the jar file, listen port is invalid,
2969 * any load fails, and can't bind port.
2971 * Else main never returns, the process must be signaled
2972 * to terminate execution. Or, on Windows, use the
2973 * "File", "Exit" menu option.
2975 *********************************************************************/
2977 int real_main(int argc, const char *argv[])
2979 int main(int argc, const char *argv[])
2983 unsigned int random_seed;
2985 struct passwd *pw = NULL;
2986 struct group *grp = NULL;
2989 char *pre_chroot_nslookup_to_load_resolver = NULL;
2996 #if !defined(_WIN32)
3004 * Parse the command line arguments
3006 * XXX: simply printing usage information in case of
3007 * invalid arguments isn't particular user friendly.
3009 while (++argc_pos < argc)
3012 /* Check to see if the service must be installed or uninstalled */
3013 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3015 const char *pName = argv[argc_pos] + 9;
3018 exit( (install_service(pName)) ? 0 : 1 );
3020 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3022 const char *pName = argv[argc_pos] + 11;
3025 exit((uninstall_service(pName)) ? 0 : 1);
3027 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3029 bRunAsService = TRUE;
3030 w32_set_service_cwd();
3031 atexit(w32_service_exit_notify);
3034 #endif /* defined(_WIN32) */
3037 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3039 if (strcmp(argv[argc_pos], "--help") == 0)
3044 else if(strcmp(argv[argc_pos], "--version") == 0)
3046 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3052 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3057 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3059 if (++argc_pos == argc) usage(argv[0]);
3060 pidfile = strdup(argv[argc_pos]);
3063 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3065 if (++argc_pos == argc) usage(argv[argc_pos]);
3067 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3070 if (NULL == (grp = getgrnam(p)))
3072 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3076 if (NULL == (pw = getpwnam(argv[argc_pos])))
3078 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3081 if (p != NULL) *--p = '\0';
3084 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3086 if (++argc_pos == argc) usage(argv[0]);
3087 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3090 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3094 #endif /* defined(unix) */
3096 else if (argc_pos + 1 != argc)
3099 * This is neither the last command line
3100 * option, nor was it recognized before,
3101 * therefore it must be invalid.
3107 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3109 configfile = argv[argc_pos];
3112 } /* -END- while (more arguments) */
3115 if ( *configfile != '/' )
3117 char cwd[BUFFER_SIZE];
3119 size_t abs_file_size;
3121 /* make config-filename absolute here */
3122 if (NULL == getcwd(cwd, sizeof(cwd)))
3124 perror("failed to get current working directory");
3129 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3130 basedir = strdup(cwd);
3132 if (NULL == basedir ||
3133 NULL == (abs_file = malloc(abs_file_size)))
3135 perror("malloc failed");
3138 strlcpy(abs_file, basedir, abs_file_size);
3139 strlcat(abs_file, "/", abs_file_size );
3140 strlcat(abs_file, configfile, abs_file_size);
3141 configfile = abs_file;
3143 #endif /* defined unix */
3147 clients->next = NULL;
3149 /* XXX: factor out initialising after the next stable release. */
3152 #elif defined(_WIN32)
3156 /* Prepare mutexes if supported and necessary. */
3157 initialize_mutexes();
3159 /* Enable logging until further notice. */
3160 init_log_module(Argv[0]);
3162 random_seed = (unsigned int)time(NULL);
3164 srandom(random_seed);
3165 #elif defined (_WIN32)
3167 * See pick_from_range() in miscutil.c for details.
3169 log_error(LOG_LEVEL_INFO,
3170 "No thread-safe PRNG implemented for your platform. "
3171 "Using weak \'randomization\' factor which will "
3172 "limit the already questionable usefulness of "
3173 "header-time-randomizing actions (disabled by default).");
3176 #endif /* ifdef HAVE_RANDOM */
3179 * Unix signal handling
3181 * Catch the abort, interrupt and terminate signals for a graceful exit
3182 * Catch the hangup signal so the errlog can be reopened.
3183 * Ignore the broken pipe signals (FIXME: Why?)
3185 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3188 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3189 const int ignored_signals[] = { SIGPIPE, 0 };
3191 for (idx = 0; catched_signals[idx] != 0; idx++)
3193 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3194 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3196 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3197 #endif /* ifdef sun */
3199 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3203 for (idx = 0; ignored_signals[idx] != 0; idx++)
3205 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3207 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3212 #else /* ifdef _WIN32 */
3213 # ifdef _WIN_CONSOLE
3215 * We *are* in a windows console app.
3216 * Print a verbose messages about FAQ's and such
3218 printf("%s", win32_blurb);
3219 # endif /* def _WIN_CONSOLE */
3220 #endif /* def _WIN32 */
3223 /* Initialize the CGI subsystem */
3224 cgi_init_error_messages();
3227 * If runnig on unix and without the --nodaemon
3228 * option, become a daemon. I.e. fork, detach
3229 * from tty and get process group leadership
3242 if ( pid < 0 ) /* error */
3247 else if ( pid != 0 ) /* parent */
3252 * must check for errors
3253 * child died due to missing files aso
3256 wpid = waitpid( pid, &status, WNOHANG );
3265 /* Should be more portable, but not as well tested */
3270 #else /* ndef __FreeBSD__ */
3272 #endif /* ndef __FreeBSD__ */
3273 fd = open("/dev/tty", O_RDONLY);
3276 /* no error check here */
3277 ioctl( fd, TIOCNOTTY,0 );
3282 * stderr (fd 2) will be closed later on, when the
3283 * log file has been parsed.
3290 } /* -END- if (!no_daemon) */
3293 * As soon as we have written the PID file, we can switch
3294 * to the user and group ID indicated by the --user option
3300 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3302 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3308 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3310 /* Read the time zone file from /etc before doing chroot. */
3312 if (NULL != pre_chroot_nslookup_to_load_resolver
3313 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3315 /* Initialize resolver library. */
3316 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3318 if (chroot(pw->pw_dir) < 0)
3320 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3324 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3327 if (setuid(pw->pw_uid))
3329 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3333 char putenv_dummy[64];
3335 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3336 if (putenv(putenv_dummy) != 0)
3338 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3341 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3342 if (putenv(putenv_dummy) != 0)
3344 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3350 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3353 #endif /* defined unix */
3356 /* This will be FALSE unless the command line specified --service
3360 /* Yup, so now we must attempt to establish a connection
3361 * with the service dispatcher. This will only work if this
3362 * process was launched by the service control manager to
3363 * actually run as a service. If this isn't the case, i've
3364 * known it take around 30 seconds or so for the call to return.
3367 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3368 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3370 /* Service has run, and at this point is now being stopped, so just return */
3375 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3377 /* An error occurred. Usually it's because --service was wrongly specified
3378 * and we were unable to connect to the Service Control Dispatcher because
3379 * it wasn't expecting us and is therefore not listening.
3381 * For now, just continue below to call the listen_loop function.
3384 #endif /* def _WIN32 */
3394 /*********************************************************************
3396 * Function : bind_port_helper
3398 * Description : Bind the listen port. Handles logging, and aborts
3402 * 1 : config = Privoxy configuration. Specifies port
3405 * Returns : Port that was opened.
3407 *********************************************************************/
3408 static jb_socket bind_port_helper(struct configuration_spec * config)
3413 if (config->haddr == NULL)
3415 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3420 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3421 config->hport, config->haddr);
3424 result = bind_port(config->haddr, config->hport, &bfd);
3431 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3432 "There may be another Privoxy or some other "
3433 "proxy running on port %d",
3434 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3435 config->hport, config->hport);
3438 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3439 "The hostname is not resolvable",
3440 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3443 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3444 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3447 /* shouldn't get here */
3448 return JB_INVALID_SOCKET;
3451 config->need_bind = 0;
3458 /* Without this simple workaround we get this compiler warning from _beginthread
3459 * warning C4028: formal parameter 1 different from declaration
3461 void w32_service_listen_loop(void *p)
3465 #endif /* def _WIN32 */
3468 /*********************************************************************
3470 * Function : listen_loop
3472 * Description : bind the listen port and enter a "FOREVER" listening loop.
3478 *********************************************************************/
3479 static void listen_loop(void)
3481 struct client_state *csp = NULL;
3483 struct configuration_spec * config;
3485 config = load_config();
3487 bfd = bind_port_helper(config);
3489 #ifdef FEATURE_GRACEFUL_TERMINATION
3490 while (!g_terminate)
3495 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3496 while (waitpid(-1, NULL, WNOHANG) > 0)
3498 /* zombie children */
3500 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3503 * Free data that was used by died threads
3509 * Re-open the errlog after HUP signal
3511 if (received_hup_signal)
3513 init_error_log(Argv[0], config->logfile);
3514 received_hup_signal = 0;
3519 #ifdef FEATURE_COOKIE_JAR
3521 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3522 * already open. I think unload_configfile() is not being run, which should do
3523 * this work. Until that can get resolved, we'll use this workaround.
3527 if (csp->config->jar)
3529 fclose(csp->config->jar);
3530 csp->config->jar = NULL;
3532 #endif /* FEATURE_COOKIE_JAR */
3533 #endif /* __OS2__ */
3535 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3537 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3541 csp->flags |= CSP_FLAG_ACTIVE;
3542 csp->sfd = JB_INVALID_SOCKET;
3544 csp->config = config = load_config();
3546 if ( config->need_bind )
3549 * Since we were listening to the "old port", we will not see
3550 * a "listen" param change until the next IJB request. So, at
3551 * least 1 more request must be made for us to find the new
3552 * setting. I am simply closing the old socket and binding the
3555 * Which-ever is correct, we will serve 1 more page via the
3556 * old settings. This should probably be a "show-proxy-args"
3557 * request. This should not be a so common of an operation
3558 * that this will hurt people's feelings.
3563 bfd = bind_port_helper(config);
3566 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3568 if (!accept_connection(csp, bfd))
3570 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3583 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3586 #ifdef FEATURE_TOGGLE
3587 if (global_toggle_state)
3588 #endif /* def FEATURE_TOGGLE */
3590 csp->flags |= CSP_FLAG_TOGGLED_ON;
3593 if (run_loader(csp))
3595 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3596 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3600 if (block_acl(NULL,csp))
3602 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3603 close_socket(csp->cfd);
3607 #endif /* def FEATURE_ACL */
3609 /* add it to the list of clients */
3610 csp->next = clients->next;
3611 clients->next = csp;
3613 if (config->multi_threaded)
3617 /* this is a switch () statment in the C preprocessor - ugh */
3618 #undef SELECTED_ONE_OPTION
3620 /* Use Pthreads in preference to native code */
3621 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3622 #define SELECTED_ONE_OPTION
3624 pthread_t the_thread;
3625 pthread_attr_t attrs;
3627 pthread_attr_init(&attrs);
3628 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3629 errno = pthread_create(&the_thread, &attrs,
3630 (void * (*)(void *))serve, csp);
3631 child_id = errno ? -1 : 0;
3632 pthread_attr_destroy(&attrs);
3636 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3637 #define SELECTED_ONE_OPTION
3638 child_id = _beginthread(
3639 (void (*)(void *))serve,
3644 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3645 #define SELECTED_ONE_OPTION
3646 child_id = _beginthread(
3647 (void(* _Optlink)(void*))serve,
3653 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3654 #define SELECTED_ONE_OPTION
3656 thread_id tid = spawn_thread
3657 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3659 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3661 child_id = (int) tid;
3670 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3671 #define SELECTED_ONE_OPTION
3672 csp->cfd = ReleaseSocket(csp->cfd, -1);
3675 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3676 NP_Output, Output(),
3677 NP_CloseOutput, FALSE,
3678 NP_Name, (ULONG)"privoxy child",
3682 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3683 NP_Output, Output(),
3684 NP_CloseOutput, FALSE,
3685 NP_Name, (ULONG)"privoxy child",
3686 NP_StackSize, 200*1024,
3692 ((struct Task *)child_id)->tc_UserData = csp;
3693 Signal((struct Task *)child_id, SIGF_SINGLE);
3698 #if !defined(SELECTED_ONE_OPTION)
3701 /* This block is only needed when using fork().
3702 * When using threads, the server thread was
3703 * created and run by the call to _beginthread().
3705 if (child_id == 0) /* child */
3708 #ifdef FEATURE_TOGGLE
3709 int inherited_toggle_state = global_toggle_state;
3710 #endif /* def FEATURE_TOGGLE */
3715 * If we've been toggled or we've blocked the request, tell Mom
3718 #ifdef FEATURE_TOGGLE
3719 if (inherited_toggle_state != global_toggle_state)
3721 rc |= RC_FLAG_TOGGLED;
3723 #endif /* def FEATURE_TOGGLE */
3725 #ifdef FEATURE_STATISTICS
3726 if (csp->flags & CSP_FLAG_REJECTED)
3728 rc |= RC_FLAG_BLOCKED;
3730 #endif /* ndef FEATURE_STATISTICS */
3734 else if (child_id > 0) /* parent */
3736 /* in a fork()'d environment, the parent's
3737 * copy of the client socket and the CSP
3741 #if !defined(_WIN32) && !defined(__CYGWIN__)
3743 wait( &child_status );
3746 * Evaluate child's return code: If the child has
3747 * - been toggled, toggle ourselves
3748 * - blocked its request, bump up the stats counter
3751 #ifdef FEATURE_TOGGLE
3752 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3754 global_toggle_state = !global_toggle_state;
3756 #endif /* def FEATURE_TOGGLE */
3758 #ifdef FEATURE_STATISTICS
3760 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3764 #endif /* def FEATURE_STATISTICS */
3766 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3767 close_socket(csp->cfd);
3768 csp->flags &= ~CSP_FLAG_ACTIVE;
3772 #undef SELECTED_ONE_OPTION
3773 /* end of cpp switch () */
3775 if (child_id < 0) /* failed */
3777 char buf[BUFFER_SIZE];
3779 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3781 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3783 write_socket(csp->cfd, buf, strlen(buf));