1 const char jcc_rcs[] = "$Id: jcc.c,v 1.161 2007/12/04 19:44:22 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) 2001-2007 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 1.161 2007/12/04 19:44:22 fabiankeil
37 * Unbreak trustfile which previously didn't work without
38 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
40 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
41 * Plug memory leak. Spotted by Valgrind, triggered by
42 * Privoxy-Regression-Test feeding proxyfuzz.py.
44 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
45 * In the HTTP snipplets, refer to the client as client.
47 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
48 * Emit a log message when activating the MS IIS5 hack.
50 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
51 * Log the "weak randomization factor" warning only
52 * once for mingw32 and provide some more details.
54 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
55 * Initialize log module after initializing mutexes, future
56 * deadlocks in that code should now work cross-platform.
58 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
59 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
60 * Reported by Bert van Leeuwen in BR#1818808.
62 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
63 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
65 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
66 * When in daemon mode, close stderr after the configuration file has been
67 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
69 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
70 * - Fix a crash when parsing invalid requests whose first header
71 * is rejected by get_header(). Regression (re?)introduced
72 * in r1.143 by yours truly.
73 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
74 * client_cookie_adder() to make sure send-vanilla-wafer can be
75 * controlled through tags (and thus regression-tested).
77 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
78 * - Move get_filter_function() from jcc.c to filters.c
79 * so the filter functions can be static.
80 * - Don't bother filtering body-less responses.
82 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
83 * Execute content filters through execute_content_filter().
85 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
86 * Initialize req to NULL to make sure it's defined if the
87 * first read_socket() call fails. Reported by icmp30.
89 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
90 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
91 * extensive comments moved to user manual.
93 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
94 * Don't crash if a broken header filter wiped out the request line.
96 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
97 * Fix byte_count calculation in case of flushes
98 * and don't parse the server headers a second time.
100 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
101 * - If there's a connection problem after we already forwarded
102 * parts of the original content, just hang up. Fixes BR#1776724.
103 * - Fix warnings about unused code on mingw32.
104 * - In case of flushes, calculate the byte count
105 * less incorrectly (I think).
107 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
108 * Add some more prototypes for static functions.
110 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
111 * Comment out request_contains_null_bytes() until it's used again.
113 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
114 * #1763173 from Stefan Huehner: s@const static@static const@
115 * and declare some more functions static.
117 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
118 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
119 * and explain why they were rejected in the first place.
120 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
121 * CONNECT requests. The request line was missing.
122 * - Add two more XXX reminders as we don't have enough already.
124 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
125 * As Hal noticed, checking dispatch_cgi() as the last cruncher
126 * looks like a bug if CGI requests are blocked unintentionally,
127 * so don't do it unless the user enabled the new config option
128 * "allow-cgi-request-crunching".
130 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
131 * - Allow to rewrite the request destination behind the client's back.
132 * - Turn the weird-looking unconditional for loop that
133 * reads the client request into a conditional while loop.
134 * Move the stuff that only runs once out of the loop.
135 * - Move parts of chat(), server_content_type() and the
136 * necessary stuff to fix BR#1750917 into get_filter_function().
138 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
139 * Temporary workaround for BR#1730105.
141 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
142 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
143 * deadlocks and crashes on OpenBSD and possibly other OS with neither
144 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
145 * Thanks to Ralf Horstmann for report and solution.
147 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
148 * Add forward-override{} to change the forwarding settings through
149 * action sections. This is mainly interesting to forward different
150 * clients differently (for example based on User-Agent or request
153 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
154 * - Let usage() mention the --chroot parameter.
155 * - Use read_socket() consistently and always leave
156 * the last buffer byte alone, even in cases where
157 * null termination (currently) doesn't matter.
159 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
160 * - Fix config file loading on Unix if no config file is specified.
161 * Since r1.97 Privoxy would always interpret the last argument as
162 * config file, even if it's a valid command line option.
163 * - Abort in case of unrecognized command line options. Closes #1719696.
164 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
165 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
167 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
168 * - Don't rerun crunchers that only depend on the request URL.
169 * - Don't count redirects and CGI requests as "blocked requests".
171 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
172 * Support crunching based on tags created by server-header taggers.
174 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
175 * Make HTTP snippets static (again). Add a Content-Type for those
176 * with content so the browser doesn't guess it based on the URL.
178 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
179 * Move crunching and request line rebuilding out of chat().
181 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
182 * Introduce tags as alternative way to specify which
183 * actions apply to a request. At the moment tags can be
184 * created based on client and server headers.
186 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
187 * Don't CLF-log CONNECT requests twice.
189 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
190 * Log the source address for ACL-related connection drops.
192 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
193 * New config option: enforce-blocks.
195 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
196 * - Move null byte check into separate function.
197 * - Don't confuse the client with error pages
198 * if a CONNECT request was already confirmed.
200 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
201 * Speed up NULL byte escaping and only log the complete
202 * NULL byte requests with header debugging enabled.
204 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
205 * Answer requests that contain NULL bytes with
206 * a custom response instead of waiting for more
207 * data until the client eventually hangs up.
209 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
210 * - Move delivery and logging of crunched responses
211 * from chat() into send_crunch_response().
212 * - Display the reason for generating http_responses.
213 * - Log the content length for LOG_LEVEL_CLF correctly
214 * (still incorrect for some fixed responses).
215 * - Reword an incorrect comment about
216 * treat-forbidden-connects-like-blocks violating
218 * - Add some log messages.
220 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
221 * Move mutex initialization into separate
222 * function and exit in case of errors.
224 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
225 * - Start to reduce chat()'s line count and move
226 * parts of it into separate functions.
227 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
229 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
230 * - Add Proxy-Agent header to HTTP snippets that are
231 * supposed to reach HTTP clients only.
232 * - Made a few CONNECT log messages more descriptive.
233 * - Catch completely empty server responses (as seen
234 * with Tor's fake ".noconnect" top level domain).
235 * - Use shiny new "forwarding-failed" template for socks errors.
237 * Revision 1.118 2007/01/07 07:43:43 joergs
238 * AmigaOS4 support added.
240 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
241 * Added config option accept-intercepted-requests
242 * and disabled it by default.
244 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
245 * Reverted parts of my last commit
246 * to keep error handling working.
248 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
249 * Fixed gcc43 conversion warnings.
251 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
252 * Fix -pedantic ISO C warning about converting
253 * from function pointer to object pointer.
255 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
256 * Silence compiler warning I introduced with my last commit.
258 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
259 * Mutex protect rand() if POSIX threading
260 * is used, warn the user if that's not possible
261 * and stop using it on _WIN32 where it could
264 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
265 * Don't prevent core dumps by catching SIGABRT.
266 * It's rude and makes debugging unreasonable painful.
268 * Revision 1.110 2006/12/13 14:52:53 etresoft
269 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
271 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
272 * Privoxy is now able to run as intercepting
273 * proxy in combination with any packet filter
274 * that does the port redirection. The destination
275 * is extracted from the "Host:" header which
276 * should be available for nearly all requests.
278 * Moved HTTP snipplets into jcc.c.
279 * Added error message for gopher proxy requests.
281 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
282 * Only unlink the pidfile if it's actually used.
284 * Change order of interception checks to make
285 * it possible to block or redirect requests for
288 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
289 * Make pthread mutex locking more generic. Instead of
290 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
291 * and use mutex locking unless there is an _r function
292 * available. Better safe than sorry.
294 * Fixes "./configure --disable-pthread" and should result
295 * in less threading-related problems on pthread-using platforms,
296 * but it still doesn't fix BR#1122404.
298 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
299 * Move pthread.h inclusion from jcc.c to jcc.h.
300 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
302 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
303 * Don't exit after receiving the second SIGHUP on Solaris.
305 * Fixes BR 1052235, but the same problem may exist on other
306 * systems. Once 3.0.6 is out we should use sigset()
307 * where available and see if it breaks anything.
309 * Revision 1.104 2006/09/23 13:26:38 roro
310 * Replace TABs by spaces in source code.
312 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
313 * Fix +redirect{}. Didn't work with -fast-redirects.
315 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
316 * Respond with 400 and a short text message
317 * if the client tries to use Privoxy as FTP proxy.
319 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
320 * Make number of retries in case of forwarded-connect problems
321 * a config file option (forwarded-connect-retries) and use 0 as
324 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
325 * Set random(3) seed.
327 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
328 * Follow the OpenBSD port's lead and protect the resolve
329 * functions on OpenBSD as well.
331 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
332 * --user fix. Only use the user as group if no group is specified.
333 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
335 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
336 * Windows service (re-)integration
340 * --install[:service_name]
341 * --uninstall[:service_name]
344 * They work as follows:
345 * --install will create a service for you and then terminate.
346 * By default the service name will be "privoxy" (without the quotes).
347 * However you can run multiple services if you wish, just by adding
348 * a colon and then a name (no spaces).
350 * --uninstall follows the exact same rules a --install.
352 * --service is used when the program is executed by the service
353 * control manager, and in normal circumstances would never be
354 * used as a command line argument.
356 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
357 * Windows service integration
359 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
360 * Incorporate Fabian Keil's patch work:
361 http://www.fabiankeil.de/sourcecode/privoxy/
363 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
364 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
365 * with what was really the latest development (the v_3_0_branch branch)
367 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
368 * Thanks to Jindrich Makovicka for a race condition fix for the log
369 * file. The race condition remains for non-pthread implementations.
370 * Reference patch #1175720.
372 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
373 * Add the ability to check jpeg images for invalid
374 * lengths of comment blocks. Defensive strategy
375 * against the exploit:
376 * Microsoft Security Bulletin MS04-028
377 * Buffer Overrun in JPEG Processing (GDI+) Could
378 * Allow Code Execution (833987)
379 * Enabled with +inspect-jpegs in actions files.
381 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
382 * - Fixed usage info for non-unix platforms
383 * - Fixed small cmdline parsing bug
385 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
386 * Diagnostics: Now preserve the returncode of pthread_create
387 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
389 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
390 * No longer ignore SIGCHLD. Fixes bug #769381
392 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
393 * Close jarfile on graceful exit, remove stray line
395 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
396 * Cosmetics: Killed a warning, a typo and an allocation left at exit
398 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
399 * No longer rely on non-POSIX.1 extensions of getcwd().
402 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
403 * Replaced setenv() by posix-compliant putenv()
404 * Thanks to Neil McCalden (nmcc AT users.sf.net).
406 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
407 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
409 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
410 * Clean-up and extension of improvements for forked mode:
411 * - Child's return code now consists of flags RC_FLAG_*
412 * - Reporting toggle to parent now properly #ifdef'ed
413 * - Children now report blocking to parent. This enables
414 * statistics in forked mode
416 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
417 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
418 * Children now report having been toggled through _exit(17), parents
419 * watch for that code and toggle themselves if found.
421 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
422 * Wrapping all *_r functions (the non-_r versions of them) with
423 * mutex semaphores for OSX. Hopefully this will take care of all
424 * of those pesky crash reports.
426 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
427 * Fixed two mostly harmless mem leaks
429 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
430 * Fix: Head of global clients list now initialized to NULL
432 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
433 * Added basic support for OPTIONS and TRACE HTTP methods:
434 * - New interceptor direct_response() added in chat().
435 * - sed() moved to earlier in the process, so that the
436 * Host: header is evaluated before actions and forwarding
439 * Revision 1.92 2002/05/08 16:00:46 oes
440 * Chat's buffer handling:
441 * - Fixed bug with unchecked out-of-mem conditions
442 * while reading client request & server headers
443 * - No longer predict if the buffer limit will be exceeded
444 * in the next read -- check add_to_iob's new
445 * return code. If buffer couldn't be extended
446 * (policy or out-of-mem) while
447 * - reading from client: abort
448 * - reading server headers: send error page
449 * - buffering server body for filter: flush,
450 * and if that fails: send error page
452 * Revision 1.91 2002/04/08 20:35:58 swa
455 * Revision 1.90 2002/04/02 14:57:28 oes
456 * Made sending wafers independent of FEATURE_COOKIE_JAR
458 * Revision 1.89 2002/03/31 17:18:59 jongfoster
459 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
461 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
462 * More compiler warning message maintenance
464 * Revision 1.87 2002/03/26 22:29:54 swa
465 * we have a new homepage!
467 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
468 * Workaround for closing the jarfile before load_config() comes around again
470 * Revision 1.85 2002/03/24 15:23:33 jongfoster
473 * Revision 1.84 2002/03/24 13:25:43 swa
474 * name change related issues
476 * Revision 1.83 2002/03/16 23:54:06 jongfoster
477 * Adding graceful termination feature, to help look for memory leaks.
478 * If you enable this (which, by design, has to be done by hand
479 * editing config.h) and then go to http://i.j.b/die, then the program
480 * will exit cleanly after the *next* request. It should free all the
481 * memory that was used.
483 * Revision 1.82 2002/03/13 00:27:05 jongfoster
486 * Revision 1.81 2002/03/12 01:42:50 oes
487 * Introduced modular filters
489 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
490 * OS/2 port maintenance:
491 * - Fixed EMX build - it had decayed a little
492 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
493 * substituted a memset for now.
495 * Revision 1.79 2002/03/09 20:03:52 jongfoster
496 * - Making various functions return int rather than size_t.
497 * (Undoing a recent change). Since size_t is unsigned on
498 * Windows, functions like read_socket that return -1 on
499 * error cannot return a size_t.
501 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
502 * crashes, and also frequently caused JB to jump to 100%
503 * CPU and stay there. (Because it thought it had just
504 * read ((unsigned)-1) == 4Gb of data...)
506 * - The signature of write_socket has changed, it now simply
507 * returns success=0/failure=nonzero.
509 * - Trying to get rid of a few warnings --with-debug on
510 * Windows, I've introduced a new type "jb_socket". This is
511 * used for the socket file descriptors. On Windows, this
512 * is SOCKET (a typedef for unsigned). Everywhere else, it's
513 * an int. The error value can't be -1 any more, so it's
514 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
515 * Windows it maps to the #define INVALID_SOCKET.)
517 * - The signature of bind_port has changed.
519 * Revision 1.78 2002/03/08 21:35:04 oes
520 * Added optional group supplement to --user option. Will now use default group of user if no group given
522 * Revision 1.77 2002/03/07 03:52:06 oes
523 * - Fixed compiler warnings etc
524 * - Improved handling of failed DNS lookups
526 * Revision 1.76 2002/03/06 22:54:35 jongfoster
527 * Automated function-comment nitpicking.
529 * Revision 1.75 2002/03/06 10:02:19 oes
530 * Fixed stupid bug when --user was not given
532 * Revision 1.74 2002/03/06 00:49:31 jongfoster
533 * Fixing warning on Windows
534 * Making #ifdefs that refer to the same variable consistently
535 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
537 * Revision 1.73 2002/03/05 23:57:30 hal9
538 * Stray character 's' on line 1618 was breaking build.
540 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
541 * - Re-enable OS/2 building after new parms were added
542 * - Fix false out of memory report when resolving CGI templates when no IP
543 * address is available of failed attempt (a la no such domain)
545 * Revision 1.71 2002/03/05 18:13:56 oes
546 * Added --user option
548 * Revision 1.70 2002/03/05 04:52:42 oes
549 * Deleted non-errlog debugging code
551 * Revision 1.69 2002/03/04 23:50:00 jongfoster
552 * Splitting off bind_port() call into bind_port_helper(), with
555 * Revision 1.68 2002/03/04 20:17:32 oes
558 * Revision 1.67 2002/03/04 18:18:57 oes
559 * - Removed _DEBUG mode
560 * - Cleand up cmdline parsing
561 * - Introduced --no-daemon, --pidfile options
562 * - Cleaned up signal handling:
563 * - Terminate cleanly on INT, TERM and ABRT
564 * - Schedule logfile for re-opening on HUP
565 * - Ignore CHLD and PIPE
566 * - Leave the rest with their default handlers
567 * - Uniform handler registration
568 * - Added usage() function
569 * - Played styleguide police
571 * Revision 1.66 2002/03/03 15:06:55 oes
572 * Re-enabled automatic config reloading
574 * Revision 1.65 2002/03/03 14:49:11 oes
575 * Fixed CLF logging: Now uses client's original HTTP request
577 * Revision 1.64 2002/03/03 09:18:03 joergs
578 * Made jumbjuster work on AmigaOS again.
580 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
581 * Clean up a little CRLF unpleasantness that suddenly appeared
583 * Revision 1.62 2002/02/20 23:17:23 jongfoster
584 * Detecting some out-of memory conditions and exiting with a log message.
586 * Revision 1.61 2002/01/17 21:01:52 jongfoster
587 * Moving all our URL and URL pattern parsing code to urlmatch.c.
589 * Revision 1.60 2001/12/30 14:07:32 steudten
590 * - Add signal handling (unix)
591 * - Add SIGHUP handler (unix)
592 * - Add creation of pidfile (unix)
593 * - Add action 'top' in rc file (RH)
594 * - Add entry 'SIGNALS' to manpage
595 * - Add exit message to logfile (unix)
597 * Revision 1.59 2001/12/13 14:07:18 oes
598 * Fixed Bug: 503 error page now sent OK
600 * Revision 1.58 2001/11/30 23:37:24 jongfoster
601 * Renaming the Win32 config file to config.txt - this is almost the
602 * same as the corresponding UNIX name "config"
604 * Revision 1.57 2001/11/16 00:47:43 jongfoster
605 * Changing the tty-disconnection code to use setsid().
607 * Revision 1.56 2001/11/13 20:20:54 jongfoster
608 * Tabs->spaces, fixing a bug with missing {} around an if()
610 * Revision 1.55 2001/11/13 20:14:53 jongfoster
611 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
613 * Revision 1.54 2001/11/07 00:03:14 steudten
614 * Give reliable return value if an error
615 * occurs not just 0 with new daemon mode.
617 * Revision 1.53 2001/11/05 21:41:43 steudten
618 * Add changes to be a real daemon just for unix os.
619 * (change cwd to /, detach from controlling tty, set
620 * process group and session leader to the own process.
622 * Add some fatal-error log message for failed malloc().
623 * Add '-d' if compiled with 'configure --with-debug' to
624 * enable debug output.
626 * Revision 1.52 2001/10/26 20:11:20 jongfoster
627 * Fixing type mismatch
629 * Revision 1.51 2001/10/26 17:38:28 oes
632 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
633 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
634 * threads to call select() simultaneously. So, it's time to do a real, live,
635 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
636 * (native). Both versions will work, but using __OS2__ offers multi-threading.
638 * Revision 1.49 2001/10/23 21:41:35 jongfoster
639 * Added call to initialize the (statically-allocated of course)
640 * "out of memory" CGI response.
642 * Revision 1.48 2001/10/10 19:56:46 jongfoster
643 * Moving some code that wasn't cookie-related out of an #ifdef
646 * Revision 1.47 2001/10/10 16:44:36 oes
647 * Added CONNECT destination port limitation check
649 * Revision 1.46 2001/10/08 15:17:41 oes
650 * Re-enabled SSL forwarding
652 * Revision 1.45 2001/10/07 15:42:11 oes
653 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
655 * Moved downgrading of the HTTP version from parse_http_request to
656 * chat(), since we can't decide if it is necessary before we have
657 * determined the actions for the URL. The HTTP command is now
658 * *always* re-built so the repairs need no longer be special-cased.
660 * filter_popups now gets a csp pointer so it can raise the new
661 * CSP_FLAG_MODIFIED flag.
665 * Added configurable size limit for the IOB. If the IOB grows so
666 * large that the next read would exceed the limit, the header
667 * is generated, and the header & unfiltered buffer are flushed
668 * to the client. Chat then continues in non-buffering,
669 * non-filtering body mode.
671 * Revision 1.44 2001/10/02 18:13:57 oes
674 * Revision 1.43 2001/10/02 15:32:13 oes
675 * Moved generation of hdr
677 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
678 * Cleaning up 2 compiler warnings on OS/2.
680 * Revision 1.41 2001/09/16 17:05:14 jongfoster
681 * Removing unused #include showarg.h
683 * Revision 1.40 2001/09/16 15:41:45 jongfoster
684 * Fixing signed/unsigned comparison warning.
686 * Revision 1.39 2001/09/16 13:21:27 jongfoster
687 * Changes to use new list functions.
689 * Revision 1.38 2001/09/16 13:01:46 jongfoster
690 * Removing redundant function call that zeroed zalloc()'d memory.
692 * Revision 1.37 2001/09/10 11:12:24 oes
693 * Deleted unused variable
695 * Revision 1.36 2001/09/10 10:56:15 oes
696 * Silenced compiler warnings
698 * Revision 1.35 2001/07/31 14:44:22 oes
699 * Deleted unused size parameter from filter_popups()
701 * Revision 1.34 2001/07/30 22:08:36 jongfoster
702 * Tidying up #defines:
703 * - All feature #defines are now of the form FEATURE_xxx
704 * - Permanently turned off WIN_GUI_EDIT
705 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
707 * Revision 1.33 2001/07/29 19:32:00 jongfoster
708 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
710 * Revision 1.32 2001/07/29 18:47:05 jongfoster
711 * Adding missing #include "loadcfg.h"
713 * Revision 1.31 2001/07/29 12:17:48 oes
714 * Applied pthread fix by Paul Lieverse
716 * Revision 1.30 2001/07/25 22:57:13 jongfoster
717 * __BEOS__ no longer overrides FEATURE_PTHREAD.
718 * This is because FEATURE_PTHREAD will soon be widely used, so I
719 * want to keep it simple.
721 * Revision 1.29 2001/07/24 12:47:06 oes
722 * Applied BeOS support update by Eugenia
724 * Revision 1.28 2001/07/23 13:26:12 oes
725 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
727 * Revision 1.27 2001/07/19 19:09:47 haroon
728 * - Added code to take care of the situation where while processing the first
729 * server response (which includes the server header), after finding the end
730 * of the headers we were not looking past the end of the headers for
731 * content modification. I enabled it for filter_popups.
732 * Someone else should look to see if other similar operations should be
733 * done to the discarded portion of the buffer.
735 * Note 2001/07/20: No, the other content modification mechanisms will process
736 * the whole iob later anyway. --oes
738 * Revision 1.26 2001/07/18 12:31:36 oes
741 * Revision 1.25 2001/07/15 19:43:49 jongfoster
742 * Supports POSIX threads.
743 * Also removed some unused #includes.
745 * Revision 1.24 2001/07/13 14:00:40 oes
746 * - Generic content modification scheme:
747 * Each feature has its own applicability flag that is set
748 * from csp->action->flags.
749 * Replaced the "filtering" int flag , by a function pointer
750 * "content_filter" to the function that will do the content
751 * modification. If it is != NULL, the document will be buffered
752 * and processed through *content_filter, which must set
753 * csp->content_length and return a modified copy of the body
754 * or return NULL (on failiure).
755 * - Changed csp->is_text to the more generic bitmap csp->content_type
756 * which can currently take the valued CT_TEXT or CT_GIF
758 * - Removed all #ifdef PCRS
760 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
761 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
762 * the `block_acl' call. This prevents a compilation error when the user
763 * does not wish to use the "ACL" feature.
765 * Revision 1.22 2001/06/29 21:45:41 oes
766 * Indentation, CRLF->LF, Tab-> Space
768 * Revision 1.21 2001/06/29 13:29:36 oes
769 * - Cleaned up, improved comments
770 * - Unified all possible interceptors (CGI,
771 * block, trust, fast_redirect) in one
772 * place, with one (CGI) answer generation
773 * mechansim. Much clearer now.
774 * - Removed the GIF image generation, which
775 * is now done in filters.c:block_url()
776 * - Made error conditions like domain lookup
777 * failiure or (various) problems while talking
778 * to the server use cgi.c:error_response()
779 * instead of generating HTML/HTTP in chat() (yuck!)
780 * - Removed logentry from cancelled commit
782 * Revision 1.20 2001/06/09 10:55:28 jongfoster
783 * Changing BUFSIZ ==> BUFFER_SIZE
785 * Revision 1.19 2001/06/07 23:12:52 jongfoster
786 * Replacing function pointer in struct gateway with a directly
787 * called function forwarded_connect().
788 * Replacing struct gateway with struct forward_spec
790 * Revision 1.18 2001/06/03 19:12:16 oes
791 * introduced new cgi handling
793 * Revision 1.17 2001/06/01 20:07:23 jongfoster
794 * Now uses action +image-blocker{} rather than config->tinygif
796 * Revision 1.16 2001/06/01 18:49:17 jongfoster
797 * Replaced "list_share" with "list" - the tiny memory gain was not
798 * worth the extra complexity.
800 * Revision 1.15 2001/05/31 21:24:47 jongfoster
801 * Changed "permission" to "action" throughout.
802 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
803 * Moved vanilla wafer check into chat(), since we must now
804 * decide whether or not to add it based on the URL.
806 * Revision 1.14 2001/05/29 20:14:01 joergs
807 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
810 * Revision 1.13 2001/05/29 09:50:24 jongfoster
811 * Unified blocklist/imagelist/permissionslist.
812 * File format is still under discussion, but the internal changes
815 * Also modified interceptor behaviour:
816 * - We now intercept all URLs beginning with one of the following
817 * prefixes (and *only* these prefixes):
819 * * http://ijbswa.sf.net/config/
820 * * http://ijbswa.sourceforge.net/config/
821 * - New interceptors "home page" - go to http://i.j.b/ to see it.
822 * - Internal changes so that intercepted and fast redirect pages
823 * are not replaced with an image.
824 * - Interceptors now have the option to send a binary page direct
825 * to the client. (i.e. ijb-send-banner uses this)
826 * - Implemented show-url-info interceptor. (Which is why I needed
827 * the above interceptors changes - a typical URL is
828 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
829 * The previous mechanism would not have intercepted that, and
830 * if it had been intercepted then it then it would have replaced
833 * Revision 1.12 2001/05/27 22:17:04 oes
835 * - re_process_buffer no longer writes the modified buffer
836 * to the client, which was very ugly. It now returns the
837 * buffer, which it is then written by chat.
839 * - content_length now adjusts the Content-Length: header
840 * for modified documents rather than crunch()ing it.
841 * (Length info in csp->content_length, which is 0 for
842 * unmodified documents)
844 * - For this to work, sed() is called twice when filtering.
846 * Revision 1.11 2001/05/26 17:27:53 jongfoster
847 * Added support for CLF and fixed LOG_LEVEL_LOG.
848 * Also did CRLF->LF fix of my previous patch.
850 * Revision 1.10 2001/05/26 15:26:15 jongfoster
851 * ACL feature now provides more security by immediately dropping
852 * connections from untrusted hosts.
854 * Revision 1.9 2001/05/26 00:28:36 jongfoster
855 * Automatic reloading of config file.
856 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
857 * Most of the global variables have been moved to a new
858 * struct configuration_spec, accessed through csp->config->globalname
859 * Most of the globals remaining are used by the Win32 GUI.
861 * Revision 1.8 2001/05/25 22:43:18 jongfoster
862 * Fixing minor memory leak and buffer overflow.
864 * Revision 1.7 2001/05/25 22:34:30 jongfoster
867 * Revision 1.6 2001/05/23 00:13:58 joergs
868 * AmigaOS support fixed.
870 * Revision 1.5 2001/05/22 18:46:04 oes
872 * - Enabled filtering banners by size rather than URL
873 * by adding patterns that replace all standard banner
874 * sizes with the "Junkbuster" gif to the re_filterfile
876 * - Enabled filtering WebBugs by providing a pattern
877 * which kills all 1x1 images
879 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
880 * which is selected by the (nonstandard and therefore
881 * capital) letter 'U' in the option string.
882 * It causes the quantifiers to be ungreedy by default.
883 * Appending a ? turns back to greedy (!).
885 * - Added a new interceptor ijb-send-banner, which
886 * sends back the "Junkbuster" gif. Without imagelist or
887 * MSIE detection support, or if tinygif = 1, or the
888 * URL isn't recognized as an imageurl, a lame HTML
889 * explanation is sent instead.
891 * - Added new feature, which permits blocking remote
892 * script redirects and firing back a local redirect
894 * The feature is conditionally compiled, i.e. it
895 * can be disabled with --disable-fast-redirects,
896 * plus it must be activated by a "fast-redirects"
897 * line in the config file, has its own log level
898 * and of course wants to be displayed by show-proxy-args
899 * Note: Boy, all the #ifdefs in 1001 locations and
900 * all the fumbling with configure.in and acconfig.h
901 * were *way* more work than the feature itself :-(
903 * - Because a generic redirect template was needed for
904 * this, tinygif = 3 now uses the same.
906 * - Moved GIFs, and other static HTTP response templates
911 * - Removed some >400 CRs again (Jon, you really worked
914 * Revision 1.4 2001/05/21 19:34:01 jongfoster
915 * Made failure to bind() a fatal error.
917 * Revision 1.3 2001/05/20 01:21:20 jongfoster
918 * Version 2.9.4 checkin.
919 * - Merged popupfile and cookiefile, and added control over PCRS
920 * filtering, in new "permissionsfile".
921 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
922 * file error you now get a message box (in the Win32 GUI) rather
923 * than the program exiting with no explanation.
924 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
926 * - Removed tabs from "config"
927 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
928 * - Bumped up version number.
930 * Revision 1.2 2001/05/17 22:34:44 oes
931 * - Added hint on GIF char array generation to jcc.c
932 * - Cleaned CRLF's from the sources and related files
933 * - Repaired logging for REF and FRC
935 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
936 * Initial import of version 2.9.3 source tree
939 *********************************************************************/
945 #include <sys/types.h>
954 # ifndef FEATURE_PTHREAD
958 # include <windows.h>
959 # include <process.h>
960 # endif /* ndef FEATURE_PTHREAD */
963 # ifndef _WIN_CONSOLE
965 # endif /* ndef _WIN_CONSOLE */
966 # include "w32svrapi.h"
968 #else /* ifndef _WIN32 */
970 # if !defined (__OS2__)
972 # include <sys/wait.h>
973 # endif /* ndef __OS2__ */
974 # include <sys/time.h>
975 # include <sys/stat.h>
976 # include <sys/ioctl.h>
979 #include <sys/termios.h>
990 # include <socket.h> /* BeOS has select() for sockets only. */
991 # include <OS.h> /* declarations for threads and stuff. */
994 # if defined(__EMX__) || defined(__OS2__)
995 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1000 #define bzero(B,N) memset(B,0x00,n)
1004 # include <select.h>
1009 #include "project.h"
1012 #include "filters.h"
1013 #include "loaders.h"
1014 #include "parsers.h"
1015 #include "killpopup.h"
1016 #include "miscutil.h"
1018 #include "jbsockets.h"
1019 #include "gateway.h"
1020 #include "actions.h"
1022 #include "loadcfg.h"
1023 #include "urlmatch.h"
1025 const char jcc_h_rcs[] = JCC_H_VERSION;
1026 const char project_h_rcs[] = PROJECT_H_VERSION;
1029 struct client_state clients[1];
1030 struct file_list files[1];
1032 #ifdef FEATURE_STATISTICS
1033 int urls_read = 0; /* total nr of urls read inc rejected */
1034 int urls_rejected = 0; /* total nr of urls rejected */
1035 #endif /* def FEATURE_STATISTICS */
1037 #ifdef FEATURE_GRACEFUL_TERMINATION
1038 int g_terminate = 0;
1041 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1042 static void sig_handler(int the_signal);
1044 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1045 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1046 static jb_err get_server_headers(struct client_state *csp);
1047 static const char *crunch_reason(const struct http_response *rsp);
1048 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
1050 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1052 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1053 static jb_err change_request_destination(struct client_state *csp);
1054 static void chat(struct client_state *csp);
1055 static void serve(struct client_state *csp);
1057 static void usage(const char *myname);
1059 static void initialize_mutexes(void);
1060 static jb_socket bind_port_helper(struct configuration_spec *config);
1061 static void listen_loop(void);
1064 void serve(struct client_state *csp);
1065 #else /* ifndef AMIGA */
1066 static void serve(struct client_state *csp);
1067 #endif /* def AMIGA */
1070 static int32 server_thread(void *data);
1071 #endif /* def __BEOS__ */
1074 #define sleep(N) Sleep(((N) * 1000))
1078 #define sleep(N) DosSleep(((N) * 100))
1081 #ifdef FEATURE_PTHREAD
1082 pthread_mutex_t log_mutex;
1083 pthread_mutex_t log_init_mutex;
1085 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1086 pthread_mutex_t resolver_mutex;
1087 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1089 #ifndef HAVE_GMTIME_R
1090 pthread_mutex_t gmtime_mutex;
1091 #endif /* ndef HAVE_GMTIME_R */
1093 #ifndef HAVE_LOCALTIME_R
1094 pthread_mutex_t localtime_mutex;
1095 #endif /* ndef HAVE_GMTIME_R */
1098 pthread_mutex_t rand_mutex;
1099 #endif /* ndef HAVE_RANDOM */
1101 #endif /* FEATURE_PTHREAD */
1103 #if defined(unix) || defined(__EMX__)
1104 const char *basedir = NULL;
1105 const char *pidfile = NULL;
1106 int received_hup_signal = 0;
1107 #endif /* defined unix */
1109 /* HTTP snipplets. */
1110 static const char CSUCCEED[] =
1111 "HTTP/1.0 200 Connection established\r\n"
1112 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1114 static const char CHEADER[] =
1115 "HTTP/1.0 400 Invalid header received from client\r\n"
1116 "Proxy-Agent: Privoxy " VERSION "\r\n"
1117 "Content-Type: text/plain\r\n"
1118 "Connection: close\r\n\r\n"
1119 "Invalid header received from client.\r\n";
1121 static const char CFORBIDDEN[] =
1122 "HTTP/1.0 403 Connection not allowable\r\n"
1123 "Proxy-Agent: Privoxy " VERSION "\r\n"
1124 "X-Hint: If you read this message interactively, then you know why this happens ,-)\r\n"
1125 "Connection: close\r\n\r\n";
1127 static const char FTP_RESPONSE[] =
1128 "HTTP/1.0 400 Invalid request received from client\r\n"
1129 "Content-Type: text/plain\r\n"
1130 "Connection: close\r\n\r\n"
1131 "Invalid request. Privoxy doesn't support FTP.\r\n";
1133 static const char GOPHER_RESPONSE[] =
1134 "HTTP/1.0 400 Invalid request received from client\r\n"
1135 "Content-Type: text/plain\r\n"
1136 "Connection: close\r\n\r\n"
1137 "Invalid request. Privoxy doesn't support gopher.\r\n";
1139 /* XXX: should be a template */
1140 static const char MISSING_DESTINATION_RESPONSE[] =
1141 "HTTP/1.0 400 Bad request received from client\r\n"
1142 "Proxy-Agent: Privoxy " VERSION "\r\n"
1143 "Content-Type: text/plain\r\n"
1144 "Connection: close\r\n\r\n"
1145 "Bad request. Privoxy was unable to extract the destination.\r\n";
1147 /* XXX: should be a template */
1148 static const char NO_SERVER_DATA_RESPONSE[] =
1149 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1150 "Proxy-Agent: Privoxy " VERSION "\r\n"
1151 "Content-Type: text/plain\r\n"
1152 "Connection: close\r\n\r\n"
1153 "Empty server or forwarder response.\r\n"
1154 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1157 /* XXX: should be a template */
1158 static const char NULL_BYTE_RESPONSE[] =
1159 "HTTP/1.0 400 Bad request received from client\r\n"
1160 "Proxy-Agent: Privoxy " VERSION "\r\n"
1161 "Content-Type: text/plain\r\n"
1162 "Connection: close\r\n\r\n"
1163 "Bad request. Null byte(s) before end of request.\r\n";
1166 /* XXX: should be a template */
1167 static const char MESSED_UP_REQUEST_RESPONSE[] =
1168 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1169 "Proxy-Agent: Privoxy " VERSION "\r\n"
1170 "Content-Type: text/plain\r\n"
1171 "Connection: close\r\n\r\n"
1172 "Bad request. Messed up with header filters.\r\n";
1174 /* A function to crunch a response */
1175 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1177 /* Crunch function flags */
1178 #define CF_NO_FLAGS 0
1179 /* Cruncher applies to forced requests as well */
1180 #define CF_IGNORE_FORCE 1
1181 /* Crunched requests are counted for the block statistics */
1182 #define CF_COUNT_AS_REJECT 2
1184 /* A crunch function and its flags */
1187 const crunch_func_ptr cruncher;
1191 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1193 /* Complete list of cruncher functions */
1194 static const struct cruncher crunchers_all[] = {
1195 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1196 { block_url, CF_COUNT_AS_REJECT },
1197 #ifdef FEATURE_TRUST
1198 { trust_url, CF_COUNT_AS_REJECT },
1199 #endif /* def FEATURE_TRUST */
1200 { redirect_url, CF_NO_FLAGS },
1201 { dispatch_cgi, CF_IGNORE_FORCE},
1205 /* Light version, used after tags are applied */
1206 static const struct cruncher crunchers_light[] = {
1207 { block_url, CF_COUNT_AS_REJECT },
1208 { redirect_url, CF_NO_FLAGS },
1213 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1214 /*********************************************************************
1216 * Function : sig_handler
1218 * Description : Signal handler for different signals.
1219 * Exit gracefully on TERM and INT
1220 * or set a flag that will cause the errlog
1221 * to be reopened by the main thread on HUP.
1224 * 1 : the_signal = the signal cause this function to call
1228 *********************************************************************/
1229 static void sig_handler(int the_signal)
1235 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1246 received_hup_signal = 1;
1251 * We shouldn't be here, unless we catch signals
1252 * in main() that we can't handle here!
1254 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1262 /*********************************************************************
1264 * Function : client_protocol_is_unsupported
1266 * Description : Checks if the client used a known unsupported
1267 * protocol and deals with it by sending an error
1271 * 1 : csp = Current client state (buffers, headers, etc...)
1272 * 2 : req = the first request line send by the client
1274 * Returns : TRUE if an error response has been generated, or
1275 * FALSE if the request doesn't look invalid.
1277 *********************************************************************/
1278 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1280 char buf[BUFFER_SIZE];
1283 * If it's a FTP or gopher request, we don't support it.
1285 * These checks are better than nothing, but they might
1286 * not work in all configurations and some clients might
1287 * have problems digesting the answer.
1289 * They should, however, never cause more problems than
1290 * Privoxy's old behaviour (returning the misleading HTML
1293 * "Could not resolve http://(ftp|gopher)://example.org").
1295 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1297 if (!strncmpic(req, "GET ftp://", 10))
1299 strlcpy(buf, FTP_RESPONSE, sizeof(buf));
1300 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1301 csp->ip_addr_str, req);
1305 strlcpy(buf, GOPHER_RESPONSE, sizeof(buf));
1306 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1307 csp->ip_addr_str, req);
1309 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1311 write_socket(csp->cfd, buf, strlen(buf));
1320 /*********************************************************************
1322 * Function : get_request_destination_elsewhere
1324 * Description : If the client's request was redirected into
1325 * Privoxy without the client's knowledge,
1326 * the request line lacks the destination host.
1328 * This function tries to get it elsewhere,
1329 * provided accept-intercepted-requests is enabled.
1331 * "Elsewhere" currently only means "Host: header",
1332 * but in the future we may ask the redirecting
1333 * packet filter to look the destination up.
1335 * If the destination stays unknown, an error
1336 * response is send to the client and headers
1337 * are freed so that chat() can return directly.
1340 * 1 : csp = Current client state (buffers, headers, etc...)
1341 * 2 : headers = a header list
1343 * Returns : JB_ERR_OK if the destination is now known, or
1344 * JB_ERR_PARSE if it isn't.
1346 *********************************************************************/
1347 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1351 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1353 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1354 " Privoxy isn't configured to accept intercepted requests.",
1355 csp->ip_addr_str, csp->http->cmd);
1356 /* XXX: Use correct size */
1357 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1358 csp->ip_addr_str, csp->http->cmd);
1360 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1361 destroy_list(headers);
1363 return JB_ERR_PARSE;
1365 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1367 /* Split the domain we just got for pattern matching */
1368 init_domain_components(csp->http);
1374 /* We can't work without destination. Go spread the news.*/
1376 req = list_to_text(headers);
1378 /* XXX: Use correct size */
1379 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1380 csp->ip_addr_str, csp->http->cmd);
1381 log_error(LOG_LEVEL_ERROR,
1382 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1383 csp->ip_addr_str, csp->http->cmd, req);
1386 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1387 destroy_list(headers);
1389 return JB_ERR_PARSE;
1392 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1393 * to get the destination IP address, use it as host directly
1394 * or do a reverse DNS lookup first.
1399 /*********************************************************************
1401 * Function : get_server_headers
1403 * Description : Parses server headers in iob and fills them
1404 * into csp->headers so that they can later be
1408 * 1 : csp = Current client state (buffers, headers, etc...)
1410 * Returns : JB_ERR_OK if everything went fine, or
1411 * JB_ERR_PARSE if the headers were incomplete.
1413 *********************************************************************/
1414 static jb_err get_server_headers(struct client_state *csp)
1416 int continue_hack_in_da_house = 0;
1419 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1424 * continue hack in da house. Ignore the ending of
1425 * this head and continue enlisting header lines.
1426 * The reason is described below.
1428 enlist(csp->headers, "");
1429 continue_hack_in_da_house = 0;
1432 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1435 * It's a bodyless continue response, don't
1436 * stop header parsing after reaching it's end.
1438 * As a result Privoxy will concatenate the
1439 * next response's head and parse and deliver
1440 * the headers as if they belonged to one request.
1442 * The client will separate them because of the
1443 * empty line between them.
1445 * XXX: What we're doing here is clearly against
1446 * the intended purpose of the continue header,
1447 * and under some conditions (HTTP/1.0 client request)
1448 * it's a standard violation.
1450 * Anyway, "sort of against the spec" is preferable
1451 * to "always getting confused by Continue responses"
1452 * (Privoxy's behaviour before this hack was added)
1454 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1455 continue_hack_in_da_house = 1;
1457 else if (*header == '\0')
1460 * If the header is empty, but the Continue hack
1461 * isn't active, we can assume that we reached the
1462 * end of the buffer before we hit the end of the
1465 * Inform the caller an let it decide how to handle it.
1467 return JB_ERR_PARSE;
1471 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1474 * XXX: Should we quit the request and return a
1475 * out of memory error page instead?
1477 log_error(LOG_LEVEL_ERROR,
1478 "Out of memory while enlisting server headers. %s lost.",
1488 /*********************************************************************
1490 * Function : crunch_reason
1492 * Description : Translates the crunch reason code into a string.
1495 * 1 : rsp = a http_response
1497 * Returns : A string with the crunch reason or an error description.
1499 *********************************************************************/
1500 static const char *crunch_reason(const struct http_response *rsp)
1502 char * reason = NULL;
1504 assert(rsp != NULL);
1507 return "Internal error while searching for crunch reason";
1510 switch (rsp->reason)
1512 case RSP_REASON_UNSUPPORTED:
1513 reason = "Unsupported HTTP feature";
1515 case RSP_REASON_BLOCKED:
1518 case RSP_REASON_UNTRUSTED:
1519 reason = "Untrusted";
1521 case RSP_REASON_REDIRECTED:
1522 reason = "Redirected";
1524 case RSP_REASON_CGI_CALL:
1525 reason = "CGI Call";
1527 case RSP_REASON_NO_SUCH_DOMAIN:
1528 reason = "DNS failure";
1530 case RSP_REASON_FORWARDING_FAILED:
1531 reason = "Forwarding failed";
1533 case RSP_REASON_CONNECT_FAILED:
1534 reason = "Connection failure";
1536 case RSP_REASON_OUT_OF_MEMORY:
1537 reason = "Out of memory (may mask other reasons)";
1540 reason = "No reason recorded";
1548 /*********************************************************************
1550 * Function : send_crunch_response
1552 * Description : Delivers already prepared response for
1553 * intercepted requests, logs the interception
1554 * and frees the response.
1557 * 1 : csp = Current client state (buffers, headers, etc...)
1558 * 1 : rsp = Fully prepared response. Will be freed on exit.
1560 * Returns : Nothing.
1562 *********************************************************************/
1563 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1565 const struct http_request *http = csp->http;
1566 char status_code[4];
1568 assert(rsp != NULL);
1569 assert(rsp->head != NULL);
1574 * Not supposed to happen. If it does
1575 * anyway, treat it as an unknown error.
1577 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1578 /* return code doesn't matter */
1583 /* If rsp is still NULL, we have serious internal problems. */
1584 log_error(LOG_LEVEL_FATAL,
1585 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1589 * Extract the status code from the actual head
1590 * that was send to the client. It is the only
1591 * way to get it right for all requests, including
1592 * the fixed ones for out-of-memory problems.
1594 * A head starts like this: 'HTTP/1.1 200...'
1598 status_code[0] = rsp->head[9];
1599 status_code[1] = rsp->head[10];
1600 status_code[2] = rsp->head[11];
1601 status_code[3] = '\0';
1603 /* Write the answer to the client */
1604 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1605 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1607 /* There is nothing we can do about it. */
1608 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1611 /* Log that the request was crunched and why. */
1612 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1613 http->hostport, http->path, crunch_reason(rsp));
1614 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1615 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1617 /* Clean up and return */
1618 if (cgi_error_memory() != rsp)
1620 free_http_response(rsp);
1627 /*********************************************************************
1629 * Function : request_contains_null_bytes
1631 * Description : Checks for NULL bytes in the request and sends
1632 * an error message to the client if any were found.
1634 * XXX: currently not used, see comment in chat().
1637 * 1 : csp = Current client state (buffers, headers, etc...)
1638 * 2 : buf = Data from the client's request to check.
1639 * 3 : len = The data length.
1641 * Returns : TRUE if the request contained one or more NULL bytes, or
1644 *********************************************************************/
1645 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1647 size_t c_len; /* Request lenght when treated as C string */
1649 c_len = strlen(buf);
1654 * Null byte(s) found. Log the request,
1655 * return an error response and hang up.
1657 size_t tmp_len = c_len;
1662 * Replace NULL byte(s) with '°' characters
1663 * so the request can be logged as string.
1664 * XXX: Is there a better replacement character?
1667 tmp_len += strlen(buf+tmp_len);
1668 } while (tmp_len < len);
1670 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1671 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1672 log_error(LOG_LEVEL_HEADER,
1673 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1675 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1677 /* XXX: Log correct size */
1678 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1688 /*********************************************************************
1690 * Function : crunch_response_triggered
1692 * Description : Checks if the request has to be crunched,
1693 * and delivers the crunch response if necessary.
1696 * 1 : csp = Current client state (buffers, headers, etc...)
1697 * 2 : crunchers = list of cruncher functions to run
1699 * Returns : TRUE if the request was answered with a crunch response
1702 *********************************************************************/
1703 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1705 struct http_response *rsp = NULL;
1706 const struct cruncher *c;
1709 * If CGI request crunching is disabled,
1710 * check the CGI dispatcher out of order to
1711 * prevent unintentional blocks or redirects.
1713 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1714 && (NULL != (rsp = dispatch_cgi(csp))))
1716 /* Deliver, log and free the interception response. */
1717 send_crunch_response(csp, rsp);
1721 for (c = crunchers; c->cruncher != NULL; c++)
1724 * Check the cruncher if either Privoxy is toggled
1725 * on and the request isn't forced, or if the cruncher
1726 * applies to forced requests as well.
1728 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1729 !(csp->flags & CSP_FLAG_FORCED)) ||
1730 (c->flags & CF_IGNORE_FORCE))
1732 rsp = c->cruncher(csp);
1735 /* Deliver, log and free the interception response. */
1736 send_crunch_response(csp, rsp);
1737 #ifdef FEATURE_STATISTICS
1738 if (c->flags & CF_COUNT_AS_REJECT)
1740 csp->flags |= CSP_FLAG_REJECTED;
1742 #endif /* def FEATURE_STATISTICS */
1753 /*********************************************************************
1755 * Function : build_request_line
1757 * Description : Builds the HTTP request line.
1759 * If a HTTP forwarder is used it expects the whole URL,
1760 * web servers only get the path.
1763 * 1 : csp = Current client state (buffers, headers, etc...)
1764 * 2 : fwd = The forwarding spec used for the request
1765 * XXX: Should use http->fwd instead.
1766 * 3 : request_line = The old request line which will be replaced.
1768 * Returns : Nothing. Terminates in case of memory problems.
1770 *********************************************************************/
1771 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1773 struct http_request *http = csp->http;
1775 assert(http->ssl == 0);
1778 * Downgrade http version from 1.1 to 1.0
1779 * if +downgrade action applies.
1781 if ( (csp->action->flags & ACTION_DOWNGRADE)
1782 && (!strcmpic(http->ver, "HTTP/1.1")))
1785 http->ver = strdup("HTTP/1.0");
1787 if (http->ver == NULL)
1789 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1794 * Rebuild the request line.
1796 freez(*request_line);
1797 *request_line = strdup(http->gpc);
1798 string_append(request_line, " ");
1800 if (fwd->forward_host)
1802 string_append(request_line, http->url);
1806 string_append(request_line, http->path);
1808 string_append(request_line, " ");
1809 string_append(request_line, http->ver);
1811 if (*request_line == NULL)
1813 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1815 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1819 /*********************************************************************
1821 * Function : change_request_destination
1823 * Description : Parse a (rewritten) request line and regenerate
1824 * the http request data.
1827 * 1 : csp = Current client state (buffers, headers, etc...)
1829 * Returns : Forwards the parse_http_request() return code.
1830 * Terminates in case of memory problems.
1832 *********************************************************************/
1833 static jb_err change_request_destination(struct client_state *csp)
1835 struct http_request *http = csp->http;
1838 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1839 free_http_request(http);
1840 err = parse_http_request(csp->headers->first->str, http, csp);
1841 if (JB_ERR_OK != err)
1843 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1844 jb_err_to_string(err));
1846 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1847 if (http->ocmd == NULL)
1849 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1856 /*********************************************************************
1860 * Description : Once a connection to the client has been accepted,
1861 * this function is called (via serve()) to handle the
1862 * main business of the communication. When this
1863 * function returns, the caller must close the client
1866 * FIXME: chat is nearly thousand lines long.
1870 * 1 : csp = Current client state (buffers, headers, etc...)
1872 * Returns : Nothing.
1874 *********************************************************************/
1875 static void chat(struct client_state *csp)
1877 char buf[BUFFER_SIZE];
1885 int ms_iis5_hack = 0;
1886 size_t byte_count = 0;
1887 int forwarded_connect_retries = 0;
1888 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1889 const struct forward_spec * fwd;
1890 struct http_request *http;
1891 int len; /* for buffer sizes (and negative error codes) */
1893 #ifdef FEATURE_KILL_POPUPS
1894 int block_popups_now = 0; /* bool, 1==currently blocking popups */
1895 #endif /* def FEATURE_KILL_POPUPS */
1897 /* Function that does the content filtering for the current request */
1898 filter_function_ptr content_filter = NULL;
1900 /* Skeleton for HTTP response, if we should intercept the request */
1901 struct http_response *rsp;
1903 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1904 struct list header_list;
1905 struct list *headers = &header_list;
1909 memset(buf, 0, sizeof(buf));
1912 * Read the client's request. Note that since we're not using select() we
1913 * could get blocked here if a client connected, then didn't say anything!
1918 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1920 if (len <= 0) break; /* error! */
1923 * If there is no memory left for buffering the
1924 * request, there is nothing we can do but hang up
1926 if (add_to_iob(csp, buf, len))
1931 req = get_header(csp);
1933 } while ((NULL != req) && ('\0' == *req));
1935 if ((NULL != req) && ('\0' != *req))
1937 /* Request received. Validate and parse it. */
1941 * XXX: Temporary disabled to prevent problems
1942 * with POST requests whose bodies are allowed to
1943 * contain NULL bytes. BR#1730105.
1945 * The main purpose of this check is to properly
1946 * log stuff like BitTorrent traffic and other junk
1947 * that hits public proxies. It's not required for
1948 * Privoxy to functions as those requests are discarded
1951 * It probably should be rewritten to only check
1952 * the head of the request. Another option would
1953 * be to let all POST requests pass, although that
1954 * may not be good enough.
1956 if (request_contains_null_bytes(csp, buf, len))
1958 /* NULL bytes found and dealt with, just hang up. */
1963 /* Does the request line look invalid? */
1964 if (client_protocol_is_unsupported(csp, req))
1967 * Yes. The request has already been
1968 * answered with a error response, the buffers
1969 * were freed and we're done with chatting.
1974 #ifdef FEATURE_FORCE_LOAD
1976 * If this request contains the FORCE_PREFIX and blocks
1977 * aren't enforced, get rid of it and set the force flag.
1979 if (strstr(req, FORCE_PREFIX))
1981 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1983 log_error(LOG_LEVEL_FORCE,
1984 "Ignored force prefix in request: \"%s\".", req);
1988 strclean(req, FORCE_PREFIX);
1989 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
1990 csp->flags |= CSP_FLAG_FORCED;
1994 #endif /* def FEATURE_FORCE_LOAD */
1995 err = parse_http_request(req, http, csp);
1996 if (JB_ERR_OK != err)
1998 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2004 if (http->cmd == NULL)
2006 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2007 /* XXX: Use correct size */
2008 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2009 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2011 free_http_request(http);
2015 /* grab the rest of the client's headers */
2019 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
2021 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2024 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2025 destroy_list(headers);
2030 * If there is no memory left for buffering the
2031 * request, there is nothing we can do but hang up
2033 if (add_to_iob(csp, buf, len))
2035 destroy_list(headers);
2041 if (p == NULL) break;
2048 if (http->host == NULL)
2051 * If we still don't know the request destination,
2052 * the request is invalid or the client uses
2053 * Privoxy without its knowledge.
2055 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2058 * Our attempts to get the request destination
2059 * elsewhere failed or Privoxy is configured
2060 * to only accept proxy requests.
2062 * An error response has already been send
2063 * and we're done here.
2070 * Determine the actions for this URL
2072 #ifdef FEATURE_TOGGLE
2073 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2075 /* Most compatible set of actions (i.e. none) */
2076 init_current_action(csp->action);
2079 #endif /* ndef FEATURE_TOGGLE */
2081 url_actions(http, csp);
2085 * Save a copy of the original request for logging
2087 http->ocmd = strdup(http->cmd);
2089 if (http->ocmd == NULL)
2091 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2094 enlist(csp->headers, http->cmd);
2096 /* Append the previously read headers */
2097 list_append_list_unique(csp->headers, headers);
2098 destroy_list(headers);
2100 err = sed(client_patterns, add_client_headers, csp);
2101 if (JB_ERR_OK != err)
2103 assert(err == JB_ERR_PARSE);
2104 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2106 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2108 /* Check request line for rewrites. */
2109 if ((NULL == csp->headers->first->str)
2110 || (strcmp(http->cmd, csp->headers->first->str) &&
2111 (JB_ERR_OK != change_request_destination(csp))))
2114 * A header filter broke the request line - bail out.
2116 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2117 /* XXX: Use correct size */
2118 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2119 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2121 free_http_request(http);
2125 /* decide how to route the HTTP request */
2126 if (NULL == (fwd = forward_url(http, csp)))
2128 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2129 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2133 * build the http request to send to the server
2134 * we have to do one of the following:
2136 * create = use the original HTTP request to create a new
2137 * HTTP request that has either the path component
2138 * without the http://domainspec (w/path) or the
2139 * full orininal URL (w/url)
2140 * Note that the path and/or the HTTP version may
2141 * have been altered by now.
2143 * connect = Open a socket to the host:port of the server
2144 * and short-circuit server and client socket.
2146 * pass = Pass the request unchanged if forwarding a CONNECT
2147 * request to a parent proxy. Note that we'll be sending
2148 * the CFAIL message ourselves if connecting to the parent
2149 * fails, but we won't send a CSUCCEED message if it works,
2150 * since that would result in a double message (ours and the
2151 * parent's). After sending the request to the parent, we simply
2154 * here's the matrix:
2157 * +--------+--------+
2159 * 0 | create | connect|
2161 * Forwarding +--------+--------+
2163 * 1 | create | pass |
2165 * +--------+--------+
2170 * Check if a CONNECT request is allowable:
2171 * In the absence of a +limit-connect action, allow only port 443.
2172 * If there is an action, allow whatever matches the specificaton.
2176 if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443)
2177 || (csp->action->flags & ACTION_LIMIT_CONNECT
2178 && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) )
2180 if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS)
2183 * The response may confuse some clients,
2184 * but makes unblocking easier.
2186 * XXX: It seems to work with all major browsers,
2187 * so we should consider returning a body by default someday ...
2189 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2190 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2191 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2193 csp->action->flags |= ACTION_BLOCK;
2198 write_socket(csp->cfd, CFORBIDDEN, strlen(CFORBIDDEN));
2199 log_error(LOG_LEVEL_INFO, "Request from %s denied. "
2200 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2201 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2203 assert(NULL != csp->http->ocmd);
2204 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 403 0", csp->ip_addr_str, csp->http->ocmd);
2206 list_remove_all(csp->headers);
2208 * XXX: For consistency we might want to log a crunch message here.
2217 freez(csp->headers->first->str);
2218 build_request_line(csp, fwd, &csp->headers->first->str);
2221 hdr = list_to_text(csp->headers);
2224 /* FIXME Should handle error properly */
2225 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2229 * We have a request. Check if one of the crunchers wants it.
2231 if (crunch_response_triggered(csp, crunchers_all))
2234 * Yes. The client got the crunch response
2235 * and we are done here after cleaning up.
2238 list_remove_all(csp->headers);
2244 * The headers can't be removed earlier because
2245 * they were still needed for the referrer check
2246 * in case of CGI crunches.
2248 * XXX: Would it be worth to move the referrer check
2249 * into client_referrer() and set a flag if it's trusted?
2251 list_remove_all(csp->headers);
2253 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2255 if (fwd->forward_host)
2257 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2258 fwd->forward_host, fwd->forward_port, http->hostport);
2262 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2265 /* here we connect to the server, gateway, or the forwarder */
2267 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2268 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2270 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2271 forwarded_connect_retries, http->hostport);
2274 if (csp->sfd == JB_INVALID_SOCKET)
2276 if (fwd->type != SOCKS_NONE)
2279 rsp = error_response(csp, "forwarding-failed", errno);
2281 else if (errno == EINVAL)
2283 rsp = error_response(csp, "no-such-domain", errno);
2287 rsp = error_response(csp, "connect-failed", errno);
2288 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2293 /* Write the answer to the client */
2296 send_crunch_response(csp, rsp);
2303 if (fwd->forward_host || (http->ssl == 0))
2305 /* write the client's (modified) header to the server
2306 * (along with anything else that may be in the buffer)
2309 if (write_socket(csp->sfd, hdr, strlen(hdr))
2310 || (flush_socket(csp->sfd, csp) < 0))
2312 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2315 rsp = error_response(csp, "connect-failed", errno);
2319 send_crunch_response(csp, rsp);
2329 * We're running an SSL tunnel and we're not forwarding,
2330 * so just send the "connect succeeded" message to the
2331 * client, flush the rest, and get out of the way.
2333 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2341 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2343 /* we're finished with the client's header */
2346 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2348 /* pass data between the client and server
2349 * until one or the other shuts down the connection.
2358 * FD_ZERO here seems to point to an errant macro which crashes.
2359 * So do this by hand for now...
2361 memset(&rfds,0x00,sizeof(fd_set));
2365 FD_SET(csp->cfd, &rfds);
2366 FD_SET(csp->sfd, &rfds);
2368 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2372 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2376 /* this is the body of the browser's request
2377 * just read it and write it.
2380 if (FD_ISSET(csp->cfd, &rfds))
2382 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2386 break; /* "game over, man" */
2389 if (write_socket(csp->sfd, buf, (size_t)len))
2391 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2398 * The server wants to talk. It could be the header or the body.
2399 * If `hdr' is null, then it's the header otherwise it's the body.
2400 * FIXME: Does `hdr' really mean `host'? No.
2404 if (FD_ISSET(csp->sfd, &rfds))
2407 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2411 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2413 if (http->ssl && (fwd->forward_host == NULL))
2416 * Just hang up. We already confirmed the client's CONNECT
2417 * request with status code 200 and unencrypted content is
2418 * no longer welcome.
2420 log_error(LOG_LEVEL_ERROR,
2421 "CONNECT already confirmed. Unable to tell the client about the problem.");
2424 else if (byte_count)
2427 * Just hang up. We already transmitted the original headers
2428 * and parts of the original content and therefore missed the
2429 * chance to send an error message (without risking data corruption).
2431 * XXX: we could retry with a fancy range request here.
2433 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2434 "Unable to tell the client about the problem.");
2438 rsp = error_response(csp, "connect-failed", errno);
2442 send_crunch_response(csp, rsp);
2448 /* Add a trailing zero. This lets filter_popups
2449 * use string operations.
2453 #ifdef FEATURE_KILL_POPUPS
2454 /* Filter the popups on this read. */
2455 if (block_popups_now)
2457 filter_popups(buf, csp);
2459 #endif /* def FEATURE_KILL_POPUPS */
2461 /* Normally, this would indicate that we've read
2462 * as much as the server has sent us and we can
2463 * close the client connection. However, Microsoft
2464 * in its wisdom has released IIS/5 with a bug that
2465 * prevents it from sending the trailing \r\n in
2466 * a 302 redirect header (and possibly other headers).
2467 * To work around this if we've haven't parsed
2468 * a full header we'll append a trailing \r\n
2469 * and see if this now generates a valid one.
2471 * This hack shouldn't have any impacts. If we've
2472 * already transmitted the header or if this is a
2473 * SSL connection, then we won't bother with this
2474 * hack. So we only work on partially received
2475 * headers. If we append a \r\n and this still
2476 * doesn't generate a valid header, then we won't
2477 * transmit anything to the client.
2482 if (server_body || http->ssl)
2485 * If we have been buffering up the document,
2486 * now is the time to apply content modification
2487 * and send the result to the client.
2491 p = execute_content_filter(csp, content_filter);
2493 * If the content filter fails, use the original
2494 * buffer and length.
2495 * (see p != NULL ? p : csp->iob->cur below)
2499 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2502 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2504 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2507 hdr = list_to_text(csp->headers);
2510 /* FIXME Should handle error properly */
2511 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2515 * Shouldn't happen because this was the second sed run
2516 * and tags are only created for the first one.
2518 assert(!crunch_response_triggered(csp, crunchers_all));
2520 if (write_socket(csp->cfd, hdr, strlen(hdr))
2521 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2523 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2533 break; /* "game over, man" */
2537 * This is NOT the body, so
2538 * Let's pretend the server just sent us a blank line.
2540 log_error(LOG_LEVEL_INFO,
2541 "Malformerd HTTP headers detected and MS IIS5 hack enabled. "
2542 "Expect an invalid response or even no response at all.");
2543 snprintf(buf, sizeof(buf), "\r\n");
2544 len = (int)strlen(buf);
2547 * Now, let the normal header parsing algorithm below do its
2548 * job. If it fails, we'll exit instead of continuing.
2555 * If this is an SSL connection or we're in the body
2556 * of the server document, just write it to the client,
2557 * unless we need to buffer the body for later content-filtering
2560 if (server_body || http->ssl)
2565 * If there is no memory left for buffering the content, or the buffer limit
2566 * has been reached, switch to non-filtering mode, i.e. make & write the
2567 * header, flush the iob and buf, and get out of the way.
2569 if (add_to_iob(csp, buf, len))
2574 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2576 hdr = list_to_text(csp->headers);
2580 * Memory is too tight to even generate the header.
2581 * Send our static "Out-of-memory" page.
2583 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2584 rsp = cgi_error_memory();
2585 send_crunch_response(csp, rsp);
2589 hdrlen = strlen(hdr);
2591 if (write_socket(csp->cfd, hdr, hdrlen)
2592 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
2593 || (write_socket(csp->cfd, buf, (size_t)len)))
2595 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2602 * Reset the byte_count to the amount of bytes
2603 * we just flushed. len will be added a few lines below,
2604 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2606 byte_count = (size_t)flushed;
2608 content_filter = NULL;
2614 if (write_socket(csp->cfd, buf, (size_t)len))
2616 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2620 byte_count += (size_t)len;
2625 /* we're still looking for the end of the
2626 * server's header ... (does that make header
2627 * parsing an "out of body experience" ?
2631 * buffer up the data we just read. If that fails,
2632 * there's little we can do but send our static
2633 * out-of-memory page.
2635 if (add_to_iob(csp, buf, len))
2637 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2638 rsp = cgi_error_memory();
2639 send_crunch_response(csp, rsp);
2644 /* Convert iob into something sed() can digest */
2645 if (JB_ERR_PARSE == get_server_headers(csp))
2649 /* Well, we tried our MS IIS/5
2650 * hack and it didn't work.
2651 * The header is incomplete
2652 * and there isn't anything
2653 * we can do about it.
2659 /* Since we have to wait for
2660 * more from the server before
2661 * we can parse the headers
2662 * we just continue here.
2668 /* Did we actually get anything? */
2669 if (NULL == csp->headers->first)
2671 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2672 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2673 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2674 free_http_request(http);
2678 /* we have now received the entire header.
2679 * filter it and send the result to the client
2681 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2683 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2685 hdr = list_to_text(csp->headers);
2688 /* FIXME Should handle error properly */
2689 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2692 if (crunch_response_triggered(csp, crunchers_light))
2695 * One of the tags created by a server-header
2696 * tagger triggered a crunch. We already
2697 * delivered the crunch response to the client
2698 * and are done here after cleaning up.
2703 /* Buffer and pcrs filter this if appropriate. */
2705 if (!http->ssl) /* We talk plaintext */
2708 #ifdef FEATURE_KILL_POPUPS
2709 /* Start blocking popups if appropriate. */
2710 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2711 (csp->action->flags & ACTION_NO_POPUPS) != 0) /* Policy allows */
2713 block_popups_now = 1;
2715 * Filter the part of the body that came in the same read
2716 * as the last headers:
2718 filter_popups(csp->iob->cur, csp);
2720 #endif /* def FEATURE_KILL_POPUPS */
2721 content_filter = get_filter_function(csp);
2724 * Only write if we're not buffering for content modification
2726 if (!content_filter)
2728 /* write the server's (modified) header to
2729 * the client (along with anything else that
2730 * may be in the buffer)
2733 if (write_socket(csp->cfd, hdr, strlen(hdr))
2734 || ((len = flush_socket(csp->cfd, csp)) < 0))
2736 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2738 /* the write failed, so don't bother
2739 * mentioning it to the client...
2740 * it probably can't hear us anyway.
2746 byte_count += (size_t)len;
2749 /* we're finished with the server's header */
2754 /* If this was a MS IIS/5 hack then it means
2755 * the server has already closed the
2756 * connection. Nothing more to read. Time
2767 return; /* huh? we should never get here */
2770 if (csp->content_length == 0)
2773 * If Privoxy didn't recalculate the
2774 * Content-Lenght, byte_count is still
2777 csp->content_length = byte_count;
2780 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2781 csp->ip_addr_str, http->ocmd, csp->content_length);
2785 /*********************************************************************
2789 * Description : This is little more than chat. We only "serve" to
2790 * to close any socket that chat may have opened.
2793 * 1 : csp = Current client state (buffers, headers, etc...)
2797 *********************************************************************/
2799 void serve(struct client_state *csp)
2800 #else /* ifndef AMIGA */
2801 static void serve(struct client_state *csp)
2802 #endif /* def AMIGA */
2805 close_socket(csp->cfd);
2807 if (csp->sfd != JB_INVALID_SOCKET)
2809 close_socket(csp->sfd);
2812 csp->flags &= ~CSP_FLAG_ACTIVE;
2818 /*********************************************************************
2820 * Function : server_thread
2822 * Description : We only exist to call `serve' in a threaded environment.
2825 * 1 : data = Current client state (buffers, headers, etc...)
2827 * Returns : Always 0.
2829 *********************************************************************/
2830 static int32 server_thread(void *data)
2832 serve((struct client_state *) data);
2840 /*********************************************************************
2844 * Description : Print usage info & exit.
2846 * Parameters : Pointer to argv[0] for identifying ourselves
2850 *********************************************************************/
2851 static void usage(const char *myname)
2853 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2857 #endif /* defined(unix) */
2860 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2861 #endif /* defined(unix) */
2862 "[--version] [configfile]\n"
2863 "Aborting\n", myname);
2868 #endif /* defined(unix) */
2871 /*********************************************************************
2873 * Function : initialize_mutexes
2875 * Description : Prepares mutexes if mutex support is available.
2879 * Returns : Void, exits in case of errors.
2881 *********************************************************************/
2882 static void initialize_mutexes(void)
2886 #ifdef FEATURE_PTHREAD
2888 * Prepare global mutex semaphores
2890 err = pthread_mutex_init(&log_mutex, 0);
2892 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2895 * XXX: The assumptions below are a bit naive
2896 * and can cause locks that aren't necessary.
2898 * For example older FreeBSD versions (< 6.x?)
2899 * have no gethostbyname_r, but gethostbyname is
2902 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2903 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2904 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2906 * XXX: should we use a single mutex for
2907 * localtime() and gmtime() as well?
2909 #ifndef HAVE_GMTIME_R
2910 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2911 #endif /* ndef HAVE_GMTIME_R */
2913 #ifndef HAVE_LOCALTIME_R
2914 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2915 #endif /* ndef HAVE_GMTIME_R */
2918 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2919 #endif /* ndef HAVE_RANDOM */
2920 #endif /* FEATURE_PTHREAD */
2923 * TODO: mutex support for mingw32 would be swell.
2928 printf("Fatal error. Mutex initialization failed: %s.\n",
2937 /*********************************************************************
2941 * Description : Load the config file and start the listen loop.
2942 * This function is a lot more *sane* with the `load_config'
2943 * and `listen_loop' functions; although it stills does
2944 * a *little* too much for my taste.
2947 * 1 : argc = Number of parameters (including $0).
2948 * 2 : argv = Array of (char *)'s to the parameters.
2950 * Returns : 1 if : can't open config file, unrecognized directive,
2951 * stats requested in multi-thread mode, can't open the
2952 * log file, can't open the jar file, listen port is invalid,
2953 * any load fails, and can't bind port.
2955 * Else main never returns, the process must be signaled
2956 * to terminate execution. Or, on Windows, use the
2957 * "File", "Exit" menu option.
2959 *********************************************************************/
2961 int real_main(int argc, const char *argv[])
2963 int main(int argc, const char *argv[])
2967 unsigned int random_seed;
2969 struct passwd *pw = NULL;
2970 struct group *grp = NULL;
2973 char *pre_chroot_nslookup_to_load_resolver = NULL;
2980 #if !defined(_WIN32)
2988 * Parse the command line arguments
2990 * XXX: simply printing usage information in case of
2991 * invalid arguments isn't particular user friendly.
2993 while (++argc_pos < argc)
2996 /* Check to see if the service must be installed or uninstalled */
2997 if (strncmp(argv[argc_pos], "--install", 9) == 0)
2999 const char *pName = argv[argc_pos] + 9;
3002 exit( (install_service(pName)) ? 0 : 1 );
3004 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3006 const char *pName = argv[argc_pos] + 11;
3009 exit((uninstall_service(pName)) ? 0 : 1);
3011 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3013 bRunAsService = TRUE;
3014 w32_set_service_cwd();
3015 atexit(w32_service_exit_notify);
3018 #endif /* defined(_WIN32) */
3021 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3023 if (strcmp(argv[argc_pos], "--help") == 0)
3028 else if(strcmp(argv[argc_pos], "--version") == 0)
3030 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3036 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3041 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3043 if (++argc_pos == argc) usage(argv[0]);
3044 pidfile = strdup(argv[argc_pos]);
3047 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3049 if (++argc_pos == argc) usage(argv[argc_pos]);
3051 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3054 if (NULL == (grp = getgrnam(p)))
3056 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3060 if (NULL == (pw = getpwnam(argv[argc_pos])))
3062 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3065 if (p != NULL) *--p = '\0';
3068 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3070 if (++argc_pos == argc) usage(argv[0]);
3071 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3074 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3078 #endif /* defined(unix) */
3080 else if (argc_pos + 1 != argc)
3083 * This is neither the last command line
3084 * option, nor was it recognized before,
3085 * therefore it must be invalid.
3091 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3093 configfile = argv[argc_pos];
3096 } /* -END- while (more arguments) */
3099 if ( *configfile != '/' )
3101 char cwd[BUFFER_SIZE];
3103 size_t abs_file_size;
3105 /* make config-filename absolute here */
3106 if (NULL == getcwd(cwd, sizeof(cwd)))
3108 perror("failed to get current working directory");
3113 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3114 basedir = strdup(cwd);
3116 if (NULL == basedir ||
3117 NULL == (abs_file = malloc(abs_file_size)))
3119 perror("malloc failed");
3122 strlcpy(abs_file, basedir, abs_file_size);
3123 strlcat(abs_file, "/", abs_file_size );
3124 strlcat(abs_file, configfile, abs_file_size);
3125 configfile = abs_file;
3127 #endif /* defined unix */
3131 clients->next = NULL;
3133 /* XXX: factor out initialising after the next stable release. */
3136 #elif defined(_WIN32)
3140 /* Prepare mutexes if supported and necessary. */
3141 initialize_mutexes();
3143 /* Enable logging until further notice. */
3144 init_log_module(Argv[0]);
3146 random_seed = (unsigned int)time(NULL);
3148 srandom(random_seed);
3149 #elif defined (_WIN32)
3151 * See pick_from_range() in miscutil.c for details.
3153 log_error(LOG_LEVEL_INFO,
3154 "No thread-safe PRNG implemented for your platform. "
3155 "Using weak \'randomization\' factor which will "
3156 "limit the already questionable usefulness of "
3157 "header-time-randomizing actions (disabled by default).");
3160 #endif /* ifdef HAVE_RANDOM */
3163 * Unix signal handling
3165 * Catch the abort, interrupt and terminate signals for a graceful exit
3166 * Catch the hangup signal so the errlog can be reopened.
3167 * Ignore the broken pipe signals (FIXME: Why?)
3169 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3172 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3173 const int ignored_signals[] = { SIGPIPE, 0 };
3175 for (idx = 0; catched_signals[idx] != 0; idx++)
3177 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3178 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3180 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3181 #endif /* ifdef sun */
3183 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3187 for (idx = 0; ignored_signals[idx] != 0; idx++)
3189 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3191 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3196 #else /* ifdef _WIN32 */
3197 # ifdef _WIN_CONSOLE
3199 * We *are* in a windows console app.
3200 * Print a verbose messages about FAQ's and such
3202 printf("%s", win32_blurb);
3203 # endif /* def _WIN_CONSOLE */
3204 #endif /* def _WIN32 */
3207 /* Initialize the CGI subsystem */
3208 cgi_init_error_messages();
3211 * If runnig on unix and without the --nodaemon
3212 * option, become a daemon. I.e. fork, detach
3213 * from tty and get process group leadership
3226 if ( pid < 0 ) /* error */
3231 else if ( pid != 0 ) /* parent */
3236 * must check for errors
3237 * child died due to missing files aso
3240 wpid = waitpid( pid, &status, WNOHANG );
3249 /* Should be more portable, but not as well tested */
3254 #else /* ndef __FreeBSD__ */
3256 #endif /* ndef __FreeBSD__ */
3257 fd = open("/dev/tty", O_RDONLY);
3260 /* no error check here */
3261 ioctl( fd, TIOCNOTTY,0 );
3266 * stderr (fd 2) will be closed later on, when the
3267 * log file has been parsed.
3274 } /* -END- if (!no_daemon) */
3277 * As soon as we have written the PID file, we can switch
3278 * to the user and group ID indicated by the --user option
3284 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3286 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3292 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3294 /* Read the time zone file from /etc before doing chroot. */
3296 if (NULL != pre_chroot_nslookup_to_load_resolver
3297 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3299 /* Initialize resolver library. */
3300 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3302 if (chroot(pw->pw_dir) < 0)
3304 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3308 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3311 if (setuid(pw->pw_uid))
3313 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3317 char putenv_dummy[64];
3319 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3320 if (putenv(putenv_dummy) != 0)
3322 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3325 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3326 if (putenv(putenv_dummy) != 0)
3328 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3334 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3337 #endif /* defined unix */
3340 /* This will be FALSE unless the command line specified --service
3344 /* Yup, so now we must attempt to establish a connection
3345 * with the service dispatcher. This will only work if this
3346 * process was launched by the service control manager to
3347 * actually run as a service. If this isn't the case, i've
3348 * known it take around 30 seconds or so for the call to return.
3351 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3352 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3354 /* Service has run, and at this point is now being stopped, so just return */
3359 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3361 /* An error occurred. Usually it's because --service was wrongly specified
3362 * and we were unable to connect to the Service Control Dispatcher because
3363 * it wasn't expecting us and is therefore not listening.
3365 * For now, just continue below to call the listen_loop function.
3368 #endif /* def _WIN32 */
3378 /*********************************************************************
3380 * Function : bind_port_helper
3382 * Description : Bind the listen port. Handles logging, and aborts
3386 * 1 : config = Privoxy configuration. Specifies port
3389 * Returns : Port that was opened.
3391 *********************************************************************/
3392 static jb_socket bind_port_helper(struct configuration_spec * config)
3397 if ( (config->haddr != NULL)
3398 && (config->haddr[0] == '1')
3399 && (config->haddr[1] == '2')
3400 && (config->haddr[2] == '7')
3401 && (config->haddr[3] == '.') )
3403 log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only",
3406 else if (config->haddr == NULL)
3408 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3413 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3414 config->hport, config->haddr);
3417 result = bind_port(config->haddr, config->hport, &bfd);
3424 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3425 "There may be another Privoxy or some other "
3426 "proxy running on port %d",
3427 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3428 config->hport, config->hport);
3431 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3432 "The hostname is not resolvable",
3433 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3436 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3437 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3440 /* shouldn't get here */
3441 return JB_INVALID_SOCKET;
3444 config->need_bind = 0;
3451 /* Without this simple workaround we get this compiler warning from _beginthread
3452 * warning C4028: formal parameter 1 different from declaration
3454 void w32_service_listen_loop(void *p)
3458 #endif /* def _WIN32 */
3461 /*********************************************************************
3463 * Function : listen_loop
3465 * Description : bind the listen port and enter a "FOREVER" listening loop.
3471 *********************************************************************/
3472 static void listen_loop(void)
3474 struct client_state *csp = NULL;
3476 struct configuration_spec * config;
3478 config = load_config();
3480 bfd = bind_port_helper(config);
3482 #ifdef FEATURE_GRACEFUL_TERMINATION
3483 while (!g_terminate)
3488 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3489 while (waitpid(-1, NULL, WNOHANG) > 0)
3491 /* zombie children */
3493 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3496 * Free data that was used by died threads
3502 * Re-open the errlog after HUP signal
3504 if (received_hup_signal)
3506 init_error_log(Argv[0], config->logfile);
3507 received_hup_signal = 0;
3512 #ifdef FEATURE_COOKIE_JAR
3514 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3515 * already open. I think unload_configfile() is not being run, which should do
3516 * this work. Until that can get resolved, we'll use this workaround.
3520 if (csp->config->jar)
3522 fclose(csp->config->jar);
3523 csp->config->jar = NULL;
3525 #endif /* FEATURE_COOKIE_JAR */
3526 #endif /* __OS2__ */
3528 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3530 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3534 csp->flags |= CSP_FLAG_ACTIVE;
3535 csp->sfd = JB_INVALID_SOCKET;
3537 csp->config = config = load_config();
3539 if ( config->need_bind )
3542 * Since we were listening to the "old port", we will not see
3543 * a "listen" param change until the next IJB request. So, at
3544 * least 1 more request must be made for us to find the new
3545 * setting. I am simply closing the old socket and binding the
3548 * Which-ever is correct, we will serve 1 more page via the
3549 * old settings. This should probably be a "show-proxy-args"
3550 * request. This should not be a so common of an operation
3551 * that this will hurt people's feelings.
3556 bfd = bind_port_helper(config);
3559 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3561 if (!accept_connection(csp, bfd))
3563 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3576 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3579 #ifdef FEATURE_TOGGLE
3580 if (global_toggle_state)
3581 #endif /* def FEATURE_TOGGLE */
3583 csp->flags |= CSP_FLAG_TOGGLED_ON;
3586 if (run_loader(csp))
3588 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3589 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3593 if (block_acl(NULL,csp))
3595 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3596 close_socket(csp->cfd);
3600 #endif /* def FEATURE_ACL */
3602 /* add it to the list of clients */
3603 csp->next = clients->next;
3604 clients->next = csp;
3606 if (config->multi_threaded)
3610 /* this is a switch () statment in the C preprocessor - ugh */
3611 #undef SELECTED_ONE_OPTION
3613 /* Use Pthreads in preference to native code */
3614 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3615 #define SELECTED_ONE_OPTION
3617 pthread_t the_thread;
3618 pthread_attr_t attrs;
3620 pthread_attr_init(&attrs);
3621 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3622 errno = pthread_create(&the_thread, &attrs,
3623 (void * (*)(void *))serve, csp);
3624 child_id = errno ? -1 : 0;
3625 pthread_attr_destroy(&attrs);
3629 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3630 #define SELECTED_ONE_OPTION
3631 child_id = _beginthread(
3632 (void (*)(void *))serve,
3637 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3638 #define SELECTED_ONE_OPTION
3639 child_id = _beginthread(
3640 (void(* _Optlink)(void*))serve,
3646 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3647 #define SELECTED_ONE_OPTION
3649 thread_id tid = spawn_thread
3650 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3652 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3654 child_id = (int) tid;
3663 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3664 #define SELECTED_ONE_OPTION
3665 csp->cfd = ReleaseSocket(csp->cfd, -1);
3668 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3669 NP_Output, Output(),
3670 NP_CloseOutput, FALSE,
3671 NP_Name, (ULONG)"privoxy child",
3675 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3676 NP_Output, Output(),
3677 NP_CloseOutput, FALSE,
3678 NP_Name, (ULONG)"privoxy child",
3679 NP_StackSize, 200*1024,
3685 ((struct Task *)child_id)->tc_UserData = csp;
3686 Signal((struct Task *)child_id, SIGF_SINGLE);
3691 #if !defined(SELECTED_ONE_OPTION)
3694 /* This block is only needed when using fork().
3695 * When using threads, the server thread was
3696 * created and run by the call to _beginthread().
3698 if (child_id == 0) /* child */
3701 #ifdef FEATURE_TOGGLE
3702 int inherited_toggle_state = global_toggle_state;
3703 #endif /* def FEATURE_TOGGLE */
3708 * If we've been toggled or we've blocked the request, tell Mom
3711 #ifdef FEATURE_TOGGLE
3712 if (inherited_toggle_state != global_toggle_state)
3714 rc |= RC_FLAG_TOGGLED;
3716 #endif /* def FEATURE_TOGGLE */
3718 #ifdef FEATURE_STATISTICS
3719 if (csp->flags & CSP_FLAG_REJECTED)
3721 rc |= RC_FLAG_BLOCKED;
3723 #endif /* ndef FEATURE_STATISTICS */
3727 else if (child_id > 0) /* parent */
3729 /* in a fork()'d environment, the parent's
3730 * copy of the client socket and the CSP
3734 #if !defined(_WIN32) && !defined(__CYGWIN__)
3736 wait( &child_status );
3739 * Evaluate child's return code: If the child has
3740 * - been toggled, toggle ourselves
3741 * - blocked its request, bump up the stats counter
3744 #ifdef FEATURE_TOGGLE
3745 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3747 global_toggle_state = !global_toggle_state;
3749 #endif /* def FEATURE_TOGGLE */
3751 #ifdef FEATURE_STATISTICS
3753 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3757 #endif /* def FEATURE_STATISTICS */
3759 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3760 close_socket(csp->cfd);
3761 csp->flags &= ~CSP_FLAG_ACTIVE;
3765 #undef SELECTED_ONE_OPTION
3766 /* end of cpp switch () */
3768 if (child_id < 0) /* failed */
3770 char buf[BUFFER_SIZE];
3772 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3774 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3776 write_socket(csp->cfd, buf, strlen(buf));
3777 close_socket(csp->cfd);
3778 csp->flags &= ~CSP_FLAG_ACTIVE;
3789 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3791 /* Clean up. Aim: free all memory (no leaks) */
3792 #ifdef FEATURE_GRACEFUL_TERMINATION
3794 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3796 unload_current_config_file();
3797 unload_current_actions_file();
3798 unload_current_re_filterfile();
3799 #ifdef FEATURE_TRUST
3800 unload_current_trust_file();
3803 if (config->multi_threaded)
3810 } while ((clients->next != NULL) && (--i > 0));
3814 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3825 #ifdef FEATURE_COOKIE_JAR
3826 if (NULL != config->jar)
3828 fclose(config->jar);
3832 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3833 /* Cleanup - remove taskbar icon etc. */
3838 #endif /* FEATURE_GRACEFUL_TERMINATION */