1 const char jcc_rcs[] = "$Id: jcc.c,v 1.162 2007/12/06 17:54:57 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 1.162 2007/12/06 17:54:57 fabiankeil
37 * Reword NO_SERVER_DATA_RESPONSE to make it harder
38 * to misunderstand what the message is all about.
40 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
41 * Unbreak trustfile which previously didn't work without
42 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
44 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
45 * Plug memory leak. Spotted by Valgrind, triggered by
46 * Privoxy-Regression-Test feeding proxyfuzz.py.
48 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
49 * In the HTTP snipplets, refer to the client as client.
51 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
52 * Emit a log message when activating the MS IIS5 hack.
54 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
55 * Log the "weak randomization factor" warning only
56 * once for mingw32 and provide some more details.
58 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
59 * Initialize log module after initializing mutexes, future
60 * deadlocks in that code should now work cross-platform.
62 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
63 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
64 * Reported by Bert van Leeuwen in BR#1818808.
66 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
67 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
69 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
70 * When in daemon mode, close stderr after the configuration file has been
71 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
73 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
74 * - Fix a crash when parsing invalid requests whose first header
75 * is rejected by get_header(). Regression (re?)introduced
76 * in r1.143 by yours truly.
77 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
78 * client_cookie_adder() to make sure send-vanilla-wafer can be
79 * controlled through tags (and thus regression-tested).
81 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
82 * - Move get_filter_function() from jcc.c to filters.c
83 * so the filter functions can be static.
84 * - Don't bother filtering body-less responses.
86 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
87 * Execute content filters through execute_content_filter().
89 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
90 * Initialize req to NULL to make sure it's defined if the
91 * first read_socket() call fails. Reported by icmp30.
93 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
94 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
95 * extensive comments moved to user manual.
97 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
98 * Don't crash if a broken header filter wiped out the request line.
100 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
101 * Fix byte_count calculation in case of flushes
102 * and don't parse the server headers a second time.
104 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
105 * - If there's a connection problem after we already forwarded
106 * parts of the original content, just hang up. Fixes BR#1776724.
107 * - Fix warnings about unused code on mingw32.
108 * - In case of flushes, calculate the byte count
109 * less incorrectly (I think).
111 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
112 * Add some more prototypes for static functions.
114 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
115 * Comment out request_contains_null_bytes() until it's used again.
117 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
118 * #1763173 from Stefan Huehner: s@const static@static const@
119 * and declare some more functions static.
121 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
122 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
123 * and explain why they were rejected in the first place.
124 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
125 * CONNECT requests. The request line was missing.
126 * - Add two more XXX reminders as we don't have enough already.
128 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
129 * As Hal noticed, checking dispatch_cgi() as the last cruncher
130 * looks like a bug if CGI requests are blocked unintentionally,
131 * so don't do it unless the user enabled the new config option
132 * "allow-cgi-request-crunching".
134 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
135 * - Allow to rewrite the request destination behind the client's back.
136 * - Turn the weird-looking unconditional for loop that
137 * reads the client request into a conditional while loop.
138 * Move the stuff that only runs once out of the loop.
139 * - Move parts of chat(), server_content_type() and the
140 * necessary stuff to fix BR#1750917 into get_filter_function().
142 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
143 * Temporary workaround for BR#1730105.
145 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
146 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
147 * deadlocks and crashes on OpenBSD and possibly other OS with neither
148 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
149 * Thanks to Ralf Horstmann for report and solution.
151 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
152 * Add forward-override{} to change the forwarding settings through
153 * action sections. This is mainly interesting to forward different
154 * clients differently (for example based on User-Agent or request
157 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
158 * - Let usage() mention the --chroot parameter.
159 * - Use read_socket() consistently and always leave
160 * the last buffer byte alone, even in cases where
161 * null termination (currently) doesn't matter.
163 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
164 * - Fix config file loading on Unix if no config file is specified.
165 * Since r1.97 Privoxy would always interpret the last argument as
166 * config file, even if it's a valid command line option.
167 * - Abort in case of unrecognized command line options. Closes #1719696.
168 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
169 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
171 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
172 * - Don't rerun crunchers that only depend on the request URL.
173 * - Don't count redirects and CGI requests as "blocked requests".
175 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
176 * Support crunching based on tags created by server-header taggers.
178 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
179 * Make HTTP snippets static (again). Add a Content-Type for those
180 * with content so the browser doesn't guess it based on the URL.
182 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
183 * Move crunching and request line rebuilding out of chat().
185 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
186 * Introduce tags as alternative way to specify which
187 * actions apply to a request. At the moment tags can be
188 * created based on client and server headers.
190 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
191 * Don't CLF-log CONNECT requests twice.
193 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
194 * Log the source address for ACL-related connection drops.
196 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
197 * New config option: enforce-blocks.
199 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
200 * - Move null byte check into separate function.
201 * - Don't confuse the client with error pages
202 * if a CONNECT request was already confirmed.
204 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
205 * Speed up NULL byte escaping and only log the complete
206 * NULL byte requests with header debugging enabled.
208 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
209 * Answer requests that contain NULL bytes with
210 * a custom response instead of waiting for more
211 * data until the client eventually hangs up.
213 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
214 * - Move delivery and logging of crunched responses
215 * from chat() into send_crunch_response().
216 * - Display the reason for generating http_responses.
217 * - Log the content length for LOG_LEVEL_CLF correctly
218 * (still incorrect for some fixed responses).
219 * - Reword an incorrect comment about
220 * treat-forbidden-connects-like-blocks violating
222 * - Add some log messages.
224 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
225 * Move mutex initialization into separate
226 * function and exit in case of errors.
228 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
229 * - Start to reduce chat()'s line count and move
230 * parts of it into separate functions.
231 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
233 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
234 * - Add Proxy-Agent header to HTTP snippets that are
235 * supposed to reach HTTP clients only.
236 * - Made a few CONNECT log messages more descriptive.
237 * - Catch completely empty server responses (as seen
238 * with Tor's fake ".noconnect" top level domain).
239 * - Use shiny new "forwarding-failed" template for socks errors.
241 * Revision 1.118 2007/01/07 07:43:43 joergs
242 * AmigaOS4 support added.
244 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
245 * Added config option accept-intercepted-requests
246 * and disabled it by default.
248 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
249 * Reverted parts of my last commit
250 * to keep error handling working.
252 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
253 * Fixed gcc43 conversion warnings.
255 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
256 * Fix -pedantic ISO C warning about converting
257 * from function pointer to object pointer.
259 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
260 * Silence compiler warning I introduced with my last commit.
262 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
263 * Mutex protect rand() if POSIX threading
264 * is used, warn the user if that's not possible
265 * and stop using it on _WIN32 where it could
268 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
269 * Don't prevent core dumps by catching SIGABRT.
270 * It's rude and makes debugging unreasonable painful.
272 * Revision 1.110 2006/12/13 14:52:53 etresoft
273 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
275 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
276 * Privoxy is now able to run as intercepting
277 * proxy in combination with any packet filter
278 * that does the port redirection. The destination
279 * is extracted from the "Host:" header which
280 * should be available for nearly all requests.
282 * Moved HTTP snipplets into jcc.c.
283 * Added error message for gopher proxy requests.
285 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
286 * Only unlink the pidfile if it's actually used.
288 * Change order of interception checks to make
289 * it possible to block or redirect requests for
292 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
293 * Make pthread mutex locking more generic. Instead of
294 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
295 * and use mutex locking unless there is an _r function
296 * available. Better safe than sorry.
298 * Fixes "./configure --disable-pthread" and should result
299 * in less threading-related problems on pthread-using platforms,
300 * but it still doesn't fix BR#1122404.
302 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
303 * Move pthread.h inclusion from jcc.c to jcc.h.
304 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
306 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
307 * Don't exit after receiving the second SIGHUP on Solaris.
309 * Fixes BR 1052235, but the same problem may exist on other
310 * systems. Once 3.0.6 is out we should use sigset()
311 * where available and see if it breaks anything.
313 * Revision 1.104 2006/09/23 13:26:38 roro
314 * Replace TABs by spaces in source code.
316 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
317 * Fix +redirect{}. Didn't work with -fast-redirects.
319 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
320 * Respond with 400 and a short text message
321 * if the client tries to use Privoxy as FTP proxy.
323 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
324 * Make number of retries in case of forwarded-connect problems
325 * a config file option (forwarded-connect-retries) and use 0 as
328 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
329 * Set random(3) seed.
331 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
332 * Follow the OpenBSD port's lead and protect the resolve
333 * functions on OpenBSD as well.
335 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
336 * --user fix. Only use the user as group if no group is specified.
337 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
339 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
340 * Windows service (re-)integration
344 * --install[:service_name]
345 * --uninstall[:service_name]
348 * They work as follows:
349 * --install will create a service for you and then terminate.
350 * By default the service name will be "privoxy" (without the quotes).
351 * However you can run multiple services if you wish, just by adding
352 * a colon and then a name (no spaces).
354 * --uninstall follows the exact same rules a --install.
356 * --service is used when the program is executed by the service
357 * control manager, and in normal circumstances would never be
358 * used as a command line argument.
360 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
361 * Windows service integration
363 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
364 * Incorporate Fabian Keil's patch work:
365 http://www.fabiankeil.de/sourcecode/privoxy/
367 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
368 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
369 * with what was really the latest development (the v_3_0_branch branch)
371 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
372 * Thanks to Jindrich Makovicka for a race condition fix for the log
373 * file. The race condition remains for non-pthread implementations.
374 * Reference patch #1175720.
376 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
377 * Add the ability to check jpeg images for invalid
378 * lengths of comment blocks. Defensive strategy
379 * against the exploit:
380 * Microsoft Security Bulletin MS04-028
381 * Buffer Overrun in JPEG Processing (GDI+) Could
382 * Allow Code Execution (833987)
383 * Enabled with +inspect-jpegs in actions files.
385 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
386 * - Fixed usage info for non-unix platforms
387 * - Fixed small cmdline parsing bug
389 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
390 * Diagnostics: Now preserve the returncode of pthread_create
391 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
393 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
394 * No longer ignore SIGCHLD. Fixes bug #769381
396 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
397 * Close jarfile on graceful exit, remove stray line
399 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
400 * Cosmetics: Killed a warning, a typo and an allocation left at exit
402 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
403 * No longer rely on non-POSIX.1 extensions of getcwd().
406 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
407 * Replaced setenv() by posix-compliant putenv()
408 * Thanks to Neil McCalden (nmcc AT users.sf.net).
410 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
411 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
413 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
414 * Clean-up and extension of improvements for forked mode:
415 * - Child's return code now consists of flags RC_FLAG_*
416 * - Reporting toggle to parent now properly #ifdef'ed
417 * - Children now report blocking to parent. This enables
418 * statistics in forked mode
420 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
421 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
422 * Children now report having been toggled through _exit(17), parents
423 * watch for that code and toggle themselves if found.
425 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
426 * Wrapping all *_r functions (the non-_r versions of them) with
427 * mutex semaphores for OSX. Hopefully this will take care of all
428 * of those pesky crash reports.
430 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
431 * Fixed two mostly harmless mem leaks
433 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
434 * Fix: Head of global clients list now initialized to NULL
436 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
437 * Added basic support for OPTIONS and TRACE HTTP methods:
438 * - New interceptor direct_response() added in chat().
439 * - sed() moved to earlier in the process, so that the
440 * Host: header is evaluated before actions and forwarding
443 * Revision 1.92 2002/05/08 16:00:46 oes
444 * Chat's buffer handling:
445 * - Fixed bug with unchecked out-of-mem conditions
446 * while reading client request & server headers
447 * - No longer predict if the buffer limit will be exceeded
448 * in the next read -- check add_to_iob's new
449 * return code. If buffer couldn't be extended
450 * (policy or out-of-mem) while
451 * - reading from client: abort
452 * - reading server headers: send error page
453 * - buffering server body for filter: flush,
454 * and if that fails: send error page
456 * Revision 1.91 2002/04/08 20:35:58 swa
459 * Revision 1.90 2002/04/02 14:57:28 oes
460 * Made sending wafers independent of FEATURE_COOKIE_JAR
462 * Revision 1.89 2002/03/31 17:18:59 jongfoster
463 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
465 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
466 * More compiler warning message maintenance
468 * Revision 1.87 2002/03/26 22:29:54 swa
469 * we have a new homepage!
471 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
472 * Workaround for closing the jarfile before load_config() comes around again
474 * Revision 1.85 2002/03/24 15:23:33 jongfoster
477 * Revision 1.84 2002/03/24 13:25:43 swa
478 * name change related issues
480 * Revision 1.83 2002/03/16 23:54:06 jongfoster
481 * Adding graceful termination feature, to help look for memory leaks.
482 * If you enable this (which, by design, has to be done by hand
483 * editing config.h) and then go to http://i.j.b/die, then the program
484 * will exit cleanly after the *next* request. It should free all the
485 * memory that was used.
487 * Revision 1.82 2002/03/13 00:27:05 jongfoster
490 * Revision 1.81 2002/03/12 01:42:50 oes
491 * Introduced modular filters
493 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
494 * OS/2 port maintenance:
495 * - Fixed EMX build - it had decayed a little
496 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
497 * substituted a memset for now.
499 * Revision 1.79 2002/03/09 20:03:52 jongfoster
500 * - Making various functions return int rather than size_t.
501 * (Undoing a recent change). Since size_t is unsigned on
502 * Windows, functions like read_socket that return -1 on
503 * error cannot return a size_t.
505 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
506 * crashes, and also frequently caused JB to jump to 100%
507 * CPU and stay there. (Because it thought it had just
508 * read ((unsigned)-1) == 4Gb of data...)
510 * - The signature of write_socket has changed, it now simply
511 * returns success=0/failure=nonzero.
513 * - Trying to get rid of a few warnings --with-debug on
514 * Windows, I've introduced a new type "jb_socket". This is
515 * used for the socket file descriptors. On Windows, this
516 * is SOCKET (a typedef for unsigned). Everywhere else, it's
517 * an int. The error value can't be -1 any more, so it's
518 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
519 * Windows it maps to the #define INVALID_SOCKET.)
521 * - The signature of bind_port has changed.
523 * Revision 1.78 2002/03/08 21:35:04 oes
524 * Added optional group supplement to --user option. Will now use default group of user if no group given
526 * Revision 1.77 2002/03/07 03:52:06 oes
527 * - Fixed compiler warnings etc
528 * - Improved handling of failed DNS lookups
530 * Revision 1.76 2002/03/06 22:54:35 jongfoster
531 * Automated function-comment nitpicking.
533 * Revision 1.75 2002/03/06 10:02:19 oes
534 * Fixed stupid bug when --user was not given
536 * Revision 1.74 2002/03/06 00:49:31 jongfoster
537 * Fixing warning on Windows
538 * Making #ifdefs that refer to the same variable consistently
539 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
541 * Revision 1.73 2002/03/05 23:57:30 hal9
542 * Stray character 's' on line 1618 was breaking build.
544 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
545 * - Re-enable OS/2 building after new parms were added
546 * - Fix false out of memory report when resolving CGI templates when no IP
547 * address is available of failed attempt (a la no such domain)
549 * Revision 1.71 2002/03/05 18:13:56 oes
550 * Added --user option
552 * Revision 1.70 2002/03/05 04:52:42 oes
553 * Deleted non-errlog debugging code
555 * Revision 1.69 2002/03/04 23:50:00 jongfoster
556 * Splitting off bind_port() call into bind_port_helper(), with
559 * Revision 1.68 2002/03/04 20:17:32 oes
562 * Revision 1.67 2002/03/04 18:18:57 oes
563 * - Removed _DEBUG mode
564 * - Cleand up cmdline parsing
565 * - Introduced --no-daemon, --pidfile options
566 * - Cleaned up signal handling:
567 * - Terminate cleanly on INT, TERM and ABRT
568 * - Schedule logfile for re-opening on HUP
569 * - Ignore CHLD and PIPE
570 * - Leave the rest with their default handlers
571 * - Uniform handler registration
572 * - Added usage() function
573 * - Played styleguide police
575 * Revision 1.66 2002/03/03 15:06:55 oes
576 * Re-enabled automatic config reloading
578 * Revision 1.65 2002/03/03 14:49:11 oes
579 * Fixed CLF logging: Now uses client's original HTTP request
581 * Revision 1.64 2002/03/03 09:18:03 joergs
582 * Made jumbjuster work on AmigaOS again.
584 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
585 * Clean up a little CRLF unpleasantness that suddenly appeared
587 * Revision 1.62 2002/02/20 23:17:23 jongfoster
588 * Detecting some out-of memory conditions and exiting with a log message.
590 * Revision 1.61 2002/01/17 21:01:52 jongfoster
591 * Moving all our URL and URL pattern parsing code to urlmatch.c.
593 * Revision 1.60 2001/12/30 14:07:32 steudten
594 * - Add signal handling (unix)
595 * - Add SIGHUP handler (unix)
596 * - Add creation of pidfile (unix)
597 * - Add action 'top' in rc file (RH)
598 * - Add entry 'SIGNALS' to manpage
599 * - Add exit message to logfile (unix)
601 * Revision 1.59 2001/12/13 14:07:18 oes
602 * Fixed Bug: 503 error page now sent OK
604 * Revision 1.58 2001/11/30 23:37:24 jongfoster
605 * Renaming the Win32 config file to config.txt - this is almost the
606 * same as the corresponding UNIX name "config"
608 * Revision 1.57 2001/11/16 00:47:43 jongfoster
609 * Changing the tty-disconnection code to use setsid().
611 * Revision 1.56 2001/11/13 20:20:54 jongfoster
612 * Tabs->spaces, fixing a bug with missing {} around an if()
614 * Revision 1.55 2001/11/13 20:14:53 jongfoster
615 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
617 * Revision 1.54 2001/11/07 00:03:14 steudten
618 * Give reliable return value if an error
619 * occurs not just 0 with new daemon mode.
621 * Revision 1.53 2001/11/05 21:41:43 steudten
622 * Add changes to be a real daemon just for unix os.
623 * (change cwd to /, detach from controlling tty, set
624 * process group and session leader to the own process.
626 * Add some fatal-error log message for failed malloc().
627 * Add '-d' if compiled with 'configure --with-debug' to
628 * enable debug output.
630 * Revision 1.52 2001/10/26 20:11:20 jongfoster
631 * Fixing type mismatch
633 * Revision 1.51 2001/10/26 17:38:28 oes
636 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
637 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
638 * threads to call select() simultaneously. So, it's time to do a real, live,
639 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
640 * (native). Both versions will work, but using __OS2__ offers multi-threading.
642 * Revision 1.49 2001/10/23 21:41:35 jongfoster
643 * Added call to initialize the (statically-allocated of course)
644 * "out of memory" CGI response.
646 * Revision 1.48 2001/10/10 19:56:46 jongfoster
647 * Moving some code that wasn't cookie-related out of an #ifdef
650 * Revision 1.47 2001/10/10 16:44:36 oes
651 * Added CONNECT destination port limitation check
653 * Revision 1.46 2001/10/08 15:17:41 oes
654 * Re-enabled SSL forwarding
656 * Revision 1.45 2001/10/07 15:42:11 oes
657 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
659 * Moved downgrading of the HTTP version from parse_http_request to
660 * chat(), since we can't decide if it is necessary before we have
661 * determined the actions for the URL. The HTTP command is now
662 * *always* re-built so the repairs need no longer be special-cased.
664 * filter_popups now gets a csp pointer so it can raise the new
665 * CSP_FLAG_MODIFIED flag.
669 * Added configurable size limit for the IOB. If the IOB grows so
670 * large that the next read would exceed the limit, the header
671 * is generated, and the header & unfiltered buffer are flushed
672 * to the client. Chat then continues in non-buffering,
673 * non-filtering body mode.
675 * Revision 1.44 2001/10/02 18:13:57 oes
678 * Revision 1.43 2001/10/02 15:32:13 oes
679 * Moved generation of hdr
681 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
682 * Cleaning up 2 compiler warnings on OS/2.
684 * Revision 1.41 2001/09/16 17:05:14 jongfoster
685 * Removing unused #include showarg.h
687 * Revision 1.40 2001/09/16 15:41:45 jongfoster
688 * Fixing signed/unsigned comparison warning.
690 * Revision 1.39 2001/09/16 13:21:27 jongfoster
691 * Changes to use new list functions.
693 * Revision 1.38 2001/09/16 13:01:46 jongfoster
694 * Removing redundant function call that zeroed zalloc()'d memory.
696 * Revision 1.37 2001/09/10 11:12:24 oes
697 * Deleted unused variable
699 * Revision 1.36 2001/09/10 10:56:15 oes
700 * Silenced compiler warnings
702 * Revision 1.35 2001/07/31 14:44:22 oes
703 * Deleted unused size parameter from filter_popups()
705 * Revision 1.34 2001/07/30 22:08:36 jongfoster
706 * Tidying up #defines:
707 * - All feature #defines are now of the form FEATURE_xxx
708 * - Permanently turned off WIN_GUI_EDIT
709 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
711 * Revision 1.33 2001/07/29 19:32:00 jongfoster
712 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
714 * Revision 1.32 2001/07/29 18:47:05 jongfoster
715 * Adding missing #include "loadcfg.h"
717 * Revision 1.31 2001/07/29 12:17:48 oes
718 * Applied pthread fix by Paul Lieverse
720 * Revision 1.30 2001/07/25 22:57:13 jongfoster
721 * __BEOS__ no longer overrides FEATURE_PTHREAD.
722 * This is because FEATURE_PTHREAD will soon be widely used, so I
723 * want to keep it simple.
725 * Revision 1.29 2001/07/24 12:47:06 oes
726 * Applied BeOS support update by Eugenia
728 * Revision 1.28 2001/07/23 13:26:12 oes
729 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
731 * Revision 1.27 2001/07/19 19:09:47 haroon
732 * - Added code to take care of the situation where while processing the first
733 * server response (which includes the server header), after finding the end
734 * of the headers we were not looking past the end of the headers for
735 * content modification. I enabled it for filter_popups.
736 * Someone else should look to see if other similar operations should be
737 * done to the discarded portion of the buffer.
739 * Note 2001/07/20: No, the other content modification mechanisms will process
740 * the whole iob later anyway. --oes
742 * Revision 1.26 2001/07/18 12:31:36 oes
745 * Revision 1.25 2001/07/15 19:43:49 jongfoster
746 * Supports POSIX threads.
747 * Also removed some unused #includes.
749 * Revision 1.24 2001/07/13 14:00:40 oes
750 * - Generic content modification scheme:
751 * Each feature has its own applicability flag that is set
752 * from csp->action->flags.
753 * Replaced the "filtering" int flag , by a function pointer
754 * "content_filter" to the function that will do the content
755 * modification. If it is != NULL, the document will be buffered
756 * and processed through *content_filter, which must set
757 * csp->content_length and return a modified copy of the body
758 * or return NULL (on failiure).
759 * - Changed csp->is_text to the more generic bitmap csp->content_type
760 * which can currently take the valued CT_TEXT or CT_GIF
762 * - Removed all #ifdef PCRS
764 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
765 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
766 * the `block_acl' call. This prevents a compilation error when the user
767 * does not wish to use the "ACL" feature.
769 * Revision 1.22 2001/06/29 21:45:41 oes
770 * Indentation, CRLF->LF, Tab-> Space
772 * Revision 1.21 2001/06/29 13:29:36 oes
773 * - Cleaned up, improved comments
774 * - Unified all possible interceptors (CGI,
775 * block, trust, fast_redirect) in one
776 * place, with one (CGI) answer generation
777 * mechansim. Much clearer now.
778 * - Removed the GIF image generation, which
779 * is now done in filters.c:block_url()
780 * - Made error conditions like domain lookup
781 * failiure or (various) problems while talking
782 * to the server use cgi.c:error_response()
783 * instead of generating HTML/HTTP in chat() (yuck!)
784 * - Removed logentry from cancelled commit
786 * Revision 1.20 2001/06/09 10:55:28 jongfoster
787 * Changing BUFSIZ ==> BUFFER_SIZE
789 * Revision 1.19 2001/06/07 23:12:52 jongfoster
790 * Replacing function pointer in struct gateway with a directly
791 * called function forwarded_connect().
792 * Replacing struct gateway with struct forward_spec
794 * Revision 1.18 2001/06/03 19:12:16 oes
795 * introduced new cgi handling
797 * Revision 1.17 2001/06/01 20:07:23 jongfoster
798 * Now uses action +image-blocker{} rather than config->tinygif
800 * Revision 1.16 2001/06/01 18:49:17 jongfoster
801 * Replaced "list_share" with "list" - the tiny memory gain was not
802 * worth the extra complexity.
804 * Revision 1.15 2001/05/31 21:24:47 jongfoster
805 * Changed "permission" to "action" throughout.
806 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
807 * Moved vanilla wafer check into chat(), since we must now
808 * decide whether or not to add it based on the URL.
810 * Revision 1.14 2001/05/29 20:14:01 joergs
811 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
814 * Revision 1.13 2001/05/29 09:50:24 jongfoster
815 * Unified blocklist/imagelist/permissionslist.
816 * File format is still under discussion, but the internal changes
819 * Also modified interceptor behaviour:
820 * - We now intercept all URLs beginning with one of the following
821 * prefixes (and *only* these prefixes):
823 * * http://ijbswa.sf.net/config/
824 * * http://ijbswa.sourceforge.net/config/
825 * - New interceptors "home page" - go to http://i.j.b/ to see it.
826 * - Internal changes so that intercepted and fast redirect pages
827 * are not replaced with an image.
828 * - Interceptors now have the option to send a binary page direct
829 * to the client. (i.e. ijb-send-banner uses this)
830 * - Implemented show-url-info interceptor. (Which is why I needed
831 * the above interceptors changes - a typical URL is
832 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
833 * The previous mechanism would not have intercepted that, and
834 * if it had been intercepted then it then it would have replaced
837 * Revision 1.12 2001/05/27 22:17:04 oes
839 * - re_process_buffer no longer writes the modified buffer
840 * to the client, which was very ugly. It now returns the
841 * buffer, which it is then written by chat.
843 * - content_length now adjusts the Content-Length: header
844 * for modified documents rather than crunch()ing it.
845 * (Length info in csp->content_length, which is 0 for
846 * unmodified documents)
848 * - For this to work, sed() is called twice when filtering.
850 * Revision 1.11 2001/05/26 17:27:53 jongfoster
851 * Added support for CLF and fixed LOG_LEVEL_LOG.
852 * Also did CRLF->LF fix of my previous patch.
854 * Revision 1.10 2001/05/26 15:26:15 jongfoster
855 * ACL feature now provides more security by immediately dropping
856 * connections from untrusted hosts.
858 * Revision 1.9 2001/05/26 00:28:36 jongfoster
859 * Automatic reloading of config file.
860 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
861 * Most of the global variables have been moved to a new
862 * struct configuration_spec, accessed through csp->config->globalname
863 * Most of the globals remaining are used by the Win32 GUI.
865 * Revision 1.8 2001/05/25 22:43:18 jongfoster
866 * Fixing minor memory leak and buffer overflow.
868 * Revision 1.7 2001/05/25 22:34:30 jongfoster
871 * Revision 1.6 2001/05/23 00:13:58 joergs
872 * AmigaOS support fixed.
874 * Revision 1.5 2001/05/22 18:46:04 oes
876 * - Enabled filtering banners by size rather than URL
877 * by adding patterns that replace all standard banner
878 * sizes with the "Junkbuster" gif to the re_filterfile
880 * - Enabled filtering WebBugs by providing a pattern
881 * which kills all 1x1 images
883 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
884 * which is selected by the (nonstandard and therefore
885 * capital) letter 'U' in the option string.
886 * It causes the quantifiers to be ungreedy by default.
887 * Appending a ? turns back to greedy (!).
889 * - Added a new interceptor ijb-send-banner, which
890 * sends back the "Junkbuster" gif. Without imagelist or
891 * MSIE detection support, or if tinygif = 1, or the
892 * URL isn't recognized as an imageurl, a lame HTML
893 * explanation is sent instead.
895 * - Added new feature, which permits blocking remote
896 * script redirects and firing back a local redirect
898 * The feature is conditionally compiled, i.e. it
899 * can be disabled with --disable-fast-redirects,
900 * plus it must be activated by a "fast-redirects"
901 * line in the config file, has its own log level
902 * and of course wants to be displayed by show-proxy-args
903 * Note: Boy, all the #ifdefs in 1001 locations and
904 * all the fumbling with configure.in and acconfig.h
905 * were *way* more work than the feature itself :-(
907 * - Because a generic redirect template was needed for
908 * this, tinygif = 3 now uses the same.
910 * - Moved GIFs, and other static HTTP response templates
915 * - Removed some >400 CRs again (Jon, you really worked
918 * Revision 1.4 2001/05/21 19:34:01 jongfoster
919 * Made failure to bind() a fatal error.
921 * Revision 1.3 2001/05/20 01:21:20 jongfoster
922 * Version 2.9.4 checkin.
923 * - Merged popupfile and cookiefile, and added control over PCRS
924 * filtering, in new "permissionsfile".
925 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
926 * file error you now get a message box (in the Win32 GUI) rather
927 * than the program exiting with no explanation.
928 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
930 * - Removed tabs from "config"
931 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
932 * - Bumped up version number.
934 * Revision 1.2 2001/05/17 22:34:44 oes
935 * - Added hint on GIF char array generation to jcc.c
936 * - Cleaned CRLF's from the sources and related files
937 * - Repaired logging for REF and FRC
939 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
940 * Initial import of version 2.9.3 source tree
943 *********************************************************************/
949 #include <sys/types.h>
958 # ifndef FEATURE_PTHREAD
962 # include <windows.h>
963 # include <process.h>
964 # endif /* ndef FEATURE_PTHREAD */
967 # ifndef _WIN_CONSOLE
969 # endif /* ndef _WIN_CONSOLE */
970 # include "w32svrapi.h"
972 #else /* ifndef _WIN32 */
974 # if !defined (__OS2__)
976 # include <sys/wait.h>
977 # endif /* ndef __OS2__ */
978 # include <sys/time.h>
979 # include <sys/stat.h>
980 # include <sys/ioctl.h>
983 #include <sys/termios.h>
994 # include <socket.h> /* BeOS has select() for sockets only. */
995 # include <OS.h> /* declarations for threads and stuff. */
998 # if defined(__EMX__) || defined(__OS2__)
999 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1004 #define bzero(B,N) memset(B,0x00,n)
1008 # include <select.h>
1013 #include "project.h"
1016 #include "filters.h"
1017 #include "loaders.h"
1018 #include "parsers.h"
1019 #include "killpopup.h"
1020 #include "miscutil.h"
1022 #include "jbsockets.h"
1023 #include "gateway.h"
1024 #include "actions.h"
1026 #include "loadcfg.h"
1027 #include "urlmatch.h"
1029 const char jcc_h_rcs[] = JCC_H_VERSION;
1030 const char project_h_rcs[] = PROJECT_H_VERSION;
1033 struct client_state clients[1];
1034 struct file_list files[1];
1036 #ifdef FEATURE_STATISTICS
1037 int urls_read = 0; /* total nr of urls read inc rejected */
1038 int urls_rejected = 0; /* total nr of urls rejected */
1039 #endif /* def FEATURE_STATISTICS */
1041 #ifdef FEATURE_GRACEFUL_TERMINATION
1042 int g_terminate = 0;
1045 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1046 static void sig_handler(int the_signal);
1048 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1049 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1050 static jb_err get_server_headers(struct client_state *csp);
1051 static const char *crunch_reason(const struct http_response *rsp);
1052 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
1054 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1056 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1057 static jb_err change_request_destination(struct client_state *csp);
1058 static void chat(struct client_state *csp);
1059 static void serve(struct client_state *csp);
1060 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1061 static void usage(const char *myname);
1063 static void initialize_mutexes(void);
1064 static jb_socket bind_port_helper(struct configuration_spec *config);
1065 static void listen_loop(void);
1068 void serve(struct client_state *csp);
1069 #else /* ifndef AMIGA */
1070 static void serve(struct client_state *csp);
1071 #endif /* def AMIGA */
1074 static int32 server_thread(void *data);
1075 #endif /* def __BEOS__ */
1078 #define sleep(N) Sleep(((N) * 1000))
1082 #define sleep(N) DosSleep(((N) * 100))
1085 #ifdef FEATURE_PTHREAD
1086 pthread_mutex_t log_mutex;
1087 pthread_mutex_t log_init_mutex;
1089 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1090 pthread_mutex_t resolver_mutex;
1091 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1093 #ifndef HAVE_GMTIME_R
1094 pthread_mutex_t gmtime_mutex;
1095 #endif /* ndef HAVE_GMTIME_R */
1097 #ifndef HAVE_LOCALTIME_R
1098 pthread_mutex_t localtime_mutex;
1099 #endif /* ndef HAVE_GMTIME_R */
1102 pthread_mutex_t rand_mutex;
1103 #endif /* ndef HAVE_RANDOM */
1105 #endif /* FEATURE_PTHREAD */
1107 #if defined(unix) || defined(__EMX__)
1108 const char *basedir = NULL;
1109 const char *pidfile = NULL;
1110 int received_hup_signal = 0;
1111 #endif /* defined unix */
1113 /* HTTP snipplets. */
1114 static const char CSUCCEED[] =
1115 "HTTP/1.0 200 Connection established\r\n"
1116 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1118 static const char CHEADER[] =
1119 "HTTP/1.0 400 Invalid header received from client\r\n"
1120 "Proxy-Agent: Privoxy " VERSION "\r\n"
1121 "Content-Type: text/plain\r\n"
1122 "Connection: close\r\n\r\n"
1123 "Invalid header received from client.\r\n";
1125 static const char CFORBIDDEN[] =
1126 "HTTP/1.0 403 Connection not allowable\r\n"
1127 "Proxy-Agent: Privoxy " VERSION "\r\n"
1128 "X-Hint: If you read this message interactively, then you know why this happens ,-)\r\n"
1129 "Connection: close\r\n\r\n";
1131 static const char FTP_RESPONSE[] =
1132 "HTTP/1.0 400 Invalid request received from client\r\n"
1133 "Content-Type: text/plain\r\n"
1134 "Connection: close\r\n\r\n"
1135 "Invalid request. Privoxy doesn't support FTP.\r\n";
1137 static const char GOPHER_RESPONSE[] =
1138 "HTTP/1.0 400 Invalid request received from client\r\n"
1139 "Content-Type: text/plain\r\n"
1140 "Connection: close\r\n\r\n"
1141 "Invalid request. Privoxy doesn't support gopher.\r\n";
1143 /* XXX: should be a template */
1144 static const char MISSING_DESTINATION_RESPONSE[] =
1145 "HTTP/1.0 400 Bad request received from client\r\n"
1146 "Proxy-Agent: Privoxy " VERSION "\r\n"
1147 "Content-Type: text/plain\r\n"
1148 "Connection: close\r\n\r\n"
1149 "Bad request. Privoxy was unable to extract the destination.\r\n";
1151 /* XXX: should be a template */
1152 static const char NO_SERVER_DATA_RESPONSE[] =
1153 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1154 "Proxy-Agent: Privoxy " VERSION "\r\n"
1155 "Content-Type: text/plain\r\n"
1156 "Connection: close\r\n\r\n"
1157 "Empty server or forwarder response.\r\n"
1158 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1161 /* XXX: should be a template */
1162 static const char NULL_BYTE_RESPONSE[] =
1163 "HTTP/1.0 400 Bad request received from client\r\n"
1164 "Proxy-Agent: Privoxy " VERSION "\r\n"
1165 "Content-Type: text/plain\r\n"
1166 "Connection: close\r\n\r\n"
1167 "Bad request. Null byte(s) before end of request.\r\n";
1170 /* XXX: should be a template */
1171 static const char MESSED_UP_REQUEST_RESPONSE[] =
1172 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1173 "Proxy-Agent: Privoxy " VERSION "\r\n"
1174 "Content-Type: text/plain\r\n"
1175 "Connection: close\r\n\r\n"
1176 "Bad request. Messed up with header filters.\r\n";
1178 /* A function to crunch a response */
1179 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1181 /* Crunch function flags */
1182 #define CF_NO_FLAGS 0
1183 /* Cruncher applies to forced requests as well */
1184 #define CF_IGNORE_FORCE 1
1185 /* Crunched requests are counted for the block statistics */
1186 #define CF_COUNT_AS_REJECT 2
1188 /* A crunch function and its flags */
1191 const crunch_func_ptr cruncher;
1195 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1197 /* Complete list of cruncher functions */
1198 static const struct cruncher crunchers_all[] = {
1199 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1200 { block_url, CF_COUNT_AS_REJECT },
1201 #ifdef FEATURE_TRUST
1202 { trust_url, CF_COUNT_AS_REJECT },
1203 #endif /* def FEATURE_TRUST */
1204 { redirect_url, CF_NO_FLAGS },
1205 { dispatch_cgi, CF_IGNORE_FORCE},
1209 /* Light version, used after tags are applied */
1210 static const struct cruncher crunchers_light[] = {
1211 { block_url, CF_COUNT_AS_REJECT },
1212 { redirect_url, CF_NO_FLAGS },
1217 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1218 /*********************************************************************
1220 * Function : sig_handler
1222 * Description : Signal handler for different signals.
1223 * Exit gracefully on TERM and INT
1224 * or set a flag that will cause the errlog
1225 * to be reopened by the main thread on HUP.
1228 * 1 : the_signal = the signal cause this function to call
1232 *********************************************************************/
1233 static void sig_handler(int the_signal)
1239 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1250 received_hup_signal = 1;
1255 * We shouldn't be here, unless we catch signals
1256 * in main() that we can't handle here!
1258 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1266 /*********************************************************************
1268 * Function : client_protocol_is_unsupported
1270 * Description : Checks if the client used a known unsupported
1271 * protocol and deals with it by sending an error
1275 * 1 : csp = Current client state (buffers, headers, etc...)
1276 * 2 : req = the first request line send by the client
1278 * Returns : TRUE if an error response has been generated, or
1279 * FALSE if the request doesn't look invalid.
1281 *********************************************************************/
1282 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1284 char buf[BUFFER_SIZE];
1287 * If it's a FTP or gopher request, we don't support it.
1289 * These checks are better than nothing, but they might
1290 * not work in all configurations and some clients might
1291 * have problems digesting the answer.
1293 * They should, however, never cause more problems than
1294 * Privoxy's old behaviour (returning the misleading HTML
1297 * "Could not resolve http://(ftp|gopher)://example.org").
1299 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1301 if (!strncmpic(req, "GET ftp://", 10))
1303 strlcpy(buf, FTP_RESPONSE, sizeof(buf));
1304 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1305 csp->ip_addr_str, req);
1309 strlcpy(buf, GOPHER_RESPONSE, sizeof(buf));
1310 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1311 csp->ip_addr_str, req);
1313 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1315 write_socket(csp->cfd, buf, strlen(buf));
1324 /*********************************************************************
1326 * Function : get_request_destination_elsewhere
1328 * Description : If the client's request was redirected into
1329 * Privoxy without the client's knowledge,
1330 * the request line lacks the destination host.
1332 * This function tries to get it elsewhere,
1333 * provided accept-intercepted-requests is enabled.
1335 * "Elsewhere" currently only means "Host: header",
1336 * but in the future we may ask the redirecting
1337 * packet filter to look the destination up.
1339 * If the destination stays unknown, an error
1340 * response is send to the client and headers
1341 * are freed so that chat() can return directly.
1344 * 1 : csp = Current client state (buffers, headers, etc...)
1345 * 2 : headers = a header list
1347 * Returns : JB_ERR_OK if the destination is now known, or
1348 * JB_ERR_PARSE if it isn't.
1350 *********************************************************************/
1351 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1355 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1357 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1358 " Privoxy isn't configured to accept intercepted requests.",
1359 csp->ip_addr_str, csp->http->cmd);
1360 /* XXX: Use correct size */
1361 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1362 csp->ip_addr_str, csp->http->cmd);
1364 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1365 destroy_list(headers);
1367 return JB_ERR_PARSE;
1369 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1371 /* Split the domain we just got for pattern matching */
1372 init_domain_components(csp->http);
1378 /* We can't work without destination. Go spread the news.*/
1380 req = list_to_text(headers);
1382 /* XXX: Use correct size */
1383 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1384 csp->ip_addr_str, csp->http->cmd);
1385 log_error(LOG_LEVEL_ERROR,
1386 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1387 csp->ip_addr_str, csp->http->cmd, req);
1390 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1391 destroy_list(headers);
1393 return JB_ERR_PARSE;
1396 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1397 * to get the destination IP address, use it as host directly
1398 * or do a reverse DNS lookup first.
1403 /*********************************************************************
1405 * Function : get_server_headers
1407 * Description : Parses server headers in iob and fills them
1408 * into csp->headers so that they can later be
1412 * 1 : csp = Current client state (buffers, headers, etc...)
1414 * Returns : JB_ERR_OK if everything went fine, or
1415 * JB_ERR_PARSE if the headers were incomplete.
1417 *********************************************************************/
1418 static jb_err get_server_headers(struct client_state *csp)
1420 int continue_hack_in_da_house = 0;
1423 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1428 * continue hack in da house. Ignore the ending of
1429 * this head and continue enlisting header lines.
1430 * The reason is described below.
1432 enlist(csp->headers, "");
1433 continue_hack_in_da_house = 0;
1436 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1439 * It's a bodyless continue response, don't
1440 * stop header parsing after reaching it's end.
1442 * As a result Privoxy will concatenate the
1443 * next response's head and parse and deliver
1444 * the headers as if they belonged to one request.
1446 * The client will separate them because of the
1447 * empty line between them.
1449 * XXX: What we're doing here is clearly against
1450 * the intended purpose of the continue header,
1451 * and under some conditions (HTTP/1.0 client request)
1452 * it's a standard violation.
1454 * Anyway, "sort of against the spec" is preferable
1455 * to "always getting confused by Continue responses"
1456 * (Privoxy's behaviour before this hack was added)
1458 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1459 continue_hack_in_da_house = 1;
1461 else if (*header == '\0')
1464 * If the header is empty, but the Continue hack
1465 * isn't active, we can assume that we reached the
1466 * end of the buffer before we hit the end of the
1469 * Inform the caller an let it decide how to handle it.
1471 return JB_ERR_PARSE;
1475 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1478 * XXX: Should we quit the request and return a
1479 * out of memory error page instead?
1481 log_error(LOG_LEVEL_ERROR,
1482 "Out of memory while enlisting server headers. %s lost.",
1492 /*********************************************************************
1494 * Function : crunch_reason
1496 * Description : Translates the crunch reason code into a string.
1499 * 1 : rsp = a http_response
1501 * Returns : A string with the crunch reason or an error description.
1503 *********************************************************************/
1504 static const char *crunch_reason(const struct http_response *rsp)
1506 char * reason = NULL;
1508 assert(rsp != NULL);
1511 return "Internal error while searching for crunch reason";
1514 switch (rsp->reason)
1516 case RSP_REASON_UNSUPPORTED:
1517 reason = "Unsupported HTTP feature";
1519 case RSP_REASON_BLOCKED:
1522 case RSP_REASON_UNTRUSTED:
1523 reason = "Untrusted";
1525 case RSP_REASON_REDIRECTED:
1526 reason = "Redirected";
1528 case RSP_REASON_CGI_CALL:
1529 reason = "CGI Call";
1531 case RSP_REASON_NO_SUCH_DOMAIN:
1532 reason = "DNS failure";
1534 case RSP_REASON_FORWARDING_FAILED:
1535 reason = "Forwarding failed";
1537 case RSP_REASON_CONNECT_FAILED:
1538 reason = "Connection failure";
1540 case RSP_REASON_OUT_OF_MEMORY:
1541 reason = "Out of memory (may mask other reasons)";
1544 reason = "No reason recorded";
1552 /*********************************************************************
1554 * Function : send_crunch_response
1556 * Description : Delivers already prepared response for
1557 * intercepted requests, logs the interception
1558 * and frees the response.
1561 * 1 : csp = Current client state (buffers, headers, etc...)
1562 * 1 : rsp = Fully prepared response. Will be freed on exit.
1564 * Returns : Nothing.
1566 *********************************************************************/
1567 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1569 const struct http_request *http = csp->http;
1570 char status_code[4];
1572 assert(rsp != NULL);
1573 assert(rsp->head != NULL);
1578 * Not supposed to happen. If it does
1579 * anyway, treat it as an unknown error.
1581 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1582 /* return code doesn't matter */
1587 /* If rsp is still NULL, we have serious internal problems. */
1588 log_error(LOG_LEVEL_FATAL,
1589 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1593 * Extract the status code from the actual head
1594 * that was send to the client. It is the only
1595 * way to get it right for all requests, including
1596 * the fixed ones for out-of-memory problems.
1598 * A head starts like this: 'HTTP/1.1 200...'
1602 status_code[0] = rsp->head[9];
1603 status_code[1] = rsp->head[10];
1604 status_code[2] = rsp->head[11];
1605 status_code[3] = '\0';
1607 /* Write the answer to the client */
1608 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1609 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1611 /* There is nothing we can do about it. */
1612 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1615 /* Log that the request was crunched and why. */
1616 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1617 http->hostport, http->path, crunch_reason(rsp));
1618 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1619 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1621 /* Clean up and return */
1622 if (cgi_error_memory() != rsp)
1624 free_http_response(rsp);
1631 /*********************************************************************
1633 * Function : request_contains_null_bytes
1635 * Description : Checks for NULL bytes in the request and sends
1636 * an error message to the client if any were found.
1638 * XXX: currently not used, see comment in chat().
1641 * 1 : csp = Current client state (buffers, headers, etc...)
1642 * 2 : buf = Data from the client's request to check.
1643 * 3 : len = The data length.
1645 * Returns : TRUE if the request contained one or more NULL bytes, or
1648 *********************************************************************/
1649 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1651 size_t c_len; /* Request lenght when treated as C string */
1653 c_len = strlen(buf);
1658 * Null byte(s) found. Log the request,
1659 * return an error response and hang up.
1661 size_t tmp_len = c_len;
1666 * Replace NULL byte(s) with '°' characters
1667 * so the request can be logged as string.
1668 * XXX: Is there a better replacement character?
1671 tmp_len += strlen(buf+tmp_len);
1672 } while (tmp_len < len);
1674 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1675 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1676 log_error(LOG_LEVEL_HEADER,
1677 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1679 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1681 /* XXX: Log correct size */
1682 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1692 /*********************************************************************
1694 * Function : crunch_response_triggered
1696 * Description : Checks if the request has to be crunched,
1697 * and delivers the crunch response if necessary.
1700 * 1 : csp = Current client state (buffers, headers, etc...)
1701 * 2 : crunchers = list of cruncher functions to run
1703 * Returns : TRUE if the request was answered with a crunch response
1706 *********************************************************************/
1707 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1709 struct http_response *rsp = NULL;
1710 const struct cruncher *c;
1713 * If CGI request crunching is disabled,
1714 * check the CGI dispatcher out of order to
1715 * prevent unintentional blocks or redirects.
1717 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1718 && (NULL != (rsp = dispatch_cgi(csp))))
1720 /* Deliver, log and free the interception response. */
1721 send_crunch_response(csp, rsp);
1725 for (c = crunchers; c->cruncher != NULL; c++)
1728 * Check the cruncher if either Privoxy is toggled
1729 * on and the request isn't forced, or if the cruncher
1730 * applies to forced requests as well.
1732 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1733 !(csp->flags & CSP_FLAG_FORCED)) ||
1734 (c->flags & CF_IGNORE_FORCE))
1736 rsp = c->cruncher(csp);
1739 /* Deliver, log and free the interception response. */
1740 send_crunch_response(csp, rsp);
1741 #ifdef FEATURE_STATISTICS
1742 if (c->flags & CF_COUNT_AS_REJECT)
1744 csp->flags |= CSP_FLAG_REJECTED;
1746 #endif /* def FEATURE_STATISTICS */
1757 /*********************************************************************
1759 * Function : build_request_line
1761 * Description : Builds the HTTP request line.
1763 * If a HTTP forwarder is used it expects the whole URL,
1764 * web servers only get the path.
1767 * 1 : csp = Current client state (buffers, headers, etc...)
1768 * 2 : fwd = The forwarding spec used for the request
1769 * XXX: Should use http->fwd instead.
1770 * 3 : request_line = The old request line which will be replaced.
1772 * Returns : Nothing. Terminates in case of memory problems.
1774 *********************************************************************/
1775 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1777 struct http_request *http = csp->http;
1779 assert(http->ssl == 0);
1782 * Downgrade http version from 1.1 to 1.0
1783 * if +downgrade action applies.
1785 if ( (csp->action->flags & ACTION_DOWNGRADE)
1786 && (!strcmpic(http->ver, "HTTP/1.1")))
1789 http->ver = strdup("HTTP/1.0");
1791 if (http->ver == NULL)
1793 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1798 * Rebuild the request line.
1800 freez(*request_line);
1801 *request_line = strdup(http->gpc);
1802 string_append(request_line, " ");
1804 if (fwd->forward_host)
1806 string_append(request_line, http->url);
1810 string_append(request_line, http->path);
1812 string_append(request_line, " ");
1813 string_append(request_line, http->ver);
1815 if (*request_line == NULL)
1817 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1819 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1823 /*********************************************************************
1825 * Function : change_request_destination
1827 * Description : Parse a (rewritten) request line and regenerate
1828 * the http request data.
1831 * 1 : csp = Current client state (buffers, headers, etc...)
1833 * Returns : Forwards the parse_http_request() return code.
1834 * Terminates in case of memory problems.
1836 *********************************************************************/
1837 static jb_err change_request_destination(struct client_state *csp)
1839 struct http_request *http = csp->http;
1842 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1843 free_http_request(http);
1844 err = parse_http_request(csp->headers->first->str, http, csp);
1845 if (JB_ERR_OK != err)
1847 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1848 jb_err_to_string(err));
1850 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1851 if (http->ocmd == NULL)
1853 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1860 /*********************************************************************
1864 * Description : Once a connection to the client has been accepted,
1865 * this function is called (via serve()) to handle the
1866 * main business of the communication. When this
1867 * function returns, the caller must close the client
1870 * FIXME: chat is nearly thousand lines long.
1874 * 1 : csp = Current client state (buffers, headers, etc...)
1876 * Returns : Nothing.
1878 *********************************************************************/
1879 static void chat(struct client_state *csp)
1881 char buf[BUFFER_SIZE];
1889 int ms_iis5_hack = 0;
1890 size_t byte_count = 0;
1891 int forwarded_connect_retries = 0;
1892 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1893 const struct forward_spec * fwd;
1894 struct http_request *http;
1895 int len; /* for buffer sizes (and negative error codes) */
1897 #ifdef FEATURE_KILL_POPUPS
1898 int block_popups_now = 0; /* bool, 1==currently blocking popups */
1899 #endif /* def FEATURE_KILL_POPUPS */
1901 /* Function that does the content filtering for the current request */
1902 filter_function_ptr content_filter = NULL;
1904 /* Skeleton for HTTP response, if we should intercept the request */
1905 struct http_response *rsp;
1907 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1908 struct list header_list;
1909 struct list *headers = &header_list;
1913 memset(buf, 0, sizeof(buf));
1916 * Read the client's request. Note that since we're not using select() we
1917 * could get blocked here if a client connected, then didn't say anything!
1922 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1924 if (len <= 0) break; /* error! */
1927 * If there is no memory left for buffering the
1928 * request, there is nothing we can do but hang up
1930 if (add_to_iob(csp, buf, len))
1935 req = get_header(csp);
1937 } while ((NULL != req) && ('\0' == *req));
1939 if ((NULL != req) && ('\0' != *req))
1941 /* Request received. Validate and parse it. */
1945 * XXX: Temporary disabled to prevent problems
1946 * with POST requests whose bodies are allowed to
1947 * contain NULL bytes. BR#1730105.
1949 * The main purpose of this check is to properly
1950 * log stuff like BitTorrent traffic and other junk
1951 * that hits public proxies. It's not required for
1952 * Privoxy to functions as those requests are discarded
1955 * It probably should be rewritten to only check
1956 * the head of the request. Another option would
1957 * be to let all POST requests pass, although that
1958 * may not be good enough.
1960 if (request_contains_null_bytes(csp, buf, len))
1962 /* NULL bytes found and dealt with, just hang up. */
1967 /* Does the request line look invalid? */
1968 if (client_protocol_is_unsupported(csp, req))
1971 * Yes. The request has already been
1972 * answered with a error response, the buffers
1973 * were freed and we're done with chatting.
1978 #ifdef FEATURE_FORCE_LOAD
1980 * If this request contains the FORCE_PREFIX and blocks
1981 * aren't enforced, get rid of it and set the force flag.
1983 if (strstr(req, FORCE_PREFIX))
1985 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1987 log_error(LOG_LEVEL_FORCE,
1988 "Ignored force prefix in request: \"%s\".", req);
1992 strclean(req, FORCE_PREFIX);
1993 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
1994 csp->flags |= CSP_FLAG_FORCED;
1998 #endif /* def FEATURE_FORCE_LOAD */
1999 err = parse_http_request(req, http, csp);
2000 if (JB_ERR_OK != err)
2002 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2008 if (http->cmd == NULL)
2010 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2011 /* XXX: Use correct size */
2012 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2013 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2015 free_http_request(http);
2019 /* grab the rest of the client's headers */
2023 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
2025 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2028 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2029 destroy_list(headers);
2034 * If there is no memory left for buffering the
2035 * request, there is nothing we can do but hang up
2037 if (add_to_iob(csp, buf, len))
2039 destroy_list(headers);
2045 if (p == NULL) break;
2052 if (http->host == NULL)
2055 * If we still don't know the request destination,
2056 * the request is invalid or the client uses
2057 * Privoxy without its knowledge.
2059 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2062 * Our attempts to get the request destination
2063 * elsewhere failed or Privoxy is configured
2064 * to only accept proxy requests.
2066 * An error response has already been send
2067 * and we're done here.
2074 * Determine the actions for this URL
2076 #ifdef FEATURE_TOGGLE
2077 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2079 /* Most compatible set of actions (i.e. none) */
2080 init_current_action(csp->action);
2083 #endif /* ndef FEATURE_TOGGLE */
2085 url_actions(http, csp);
2089 * Save a copy of the original request for logging
2091 http->ocmd = strdup(http->cmd);
2093 if (http->ocmd == NULL)
2095 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2098 enlist(csp->headers, http->cmd);
2100 /* Append the previously read headers */
2101 list_append_list_unique(csp->headers, headers);
2102 destroy_list(headers);
2104 err = sed(client_patterns, add_client_headers, csp);
2105 if (JB_ERR_OK != err)
2107 assert(err == JB_ERR_PARSE);
2108 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2110 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2112 /* Check request line for rewrites. */
2113 if ((NULL == csp->headers->first->str)
2114 || (strcmp(http->cmd, csp->headers->first->str) &&
2115 (JB_ERR_OK != change_request_destination(csp))))
2118 * A header filter broke the request line - bail out.
2120 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2121 /* XXX: Use correct size */
2122 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2123 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2125 free_http_request(http);
2129 /* decide how to route the HTTP request */
2130 if (NULL == (fwd = forward_url(http, csp)))
2132 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2133 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2137 * build the http request to send to the server
2138 * we have to do one of the following:
2140 * create = use the original HTTP request to create a new
2141 * HTTP request that has either the path component
2142 * without the http://domainspec (w/path) or the
2143 * full orininal URL (w/url)
2144 * Note that the path and/or the HTTP version may
2145 * have been altered by now.
2147 * connect = Open a socket to the host:port of the server
2148 * and short-circuit server and client socket.
2150 * pass = Pass the request unchanged if forwarding a CONNECT
2151 * request to a parent proxy. Note that we'll be sending
2152 * the CFAIL message ourselves if connecting to the parent
2153 * fails, but we won't send a CSUCCEED message if it works,
2154 * since that would result in a double message (ours and the
2155 * parent's). After sending the request to the parent, we simply
2158 * here's the matrix:
2161 * +--------+--------+
2163 * 0 | create | connect|
2165 * Forwarding +--------+--------+
2167 * 1 | create | pass |
2169 * +--------+--------+
2174 * Check if a CONNECT request is allowable:
2175 * In the absence of a +limit-connect action, allow only port 443.
2176 * If there is an action, allow whatever matches the specificaton.
2180 if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443)
2181 || (csp->action->flags & ACTION_LIMIT_CONNECT
2182 && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) )
2184 if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS)
2187 * The response may confuse some clients,
2188 * but makes unblocking easier.
2190 * XXX: It seems to work with all major browsers,
2191 * so we should consider returning a body by default someday ...
2193 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2194 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2195 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2197 csp->action->flags |= ACTION_BLOCK;
2202 write_socket(csp->cfd, CFORBIDDEN, strlen(CFORBIDDEN));
2203 log_error(LOG_LEVEL_INFO, "Request from %s denied. "
2204 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2205 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2207 assert(NULL != csp->http->ocmd);
2208 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 403 0", csp->ip_addr_str, csp->http->ocmd);
2210 list_remove_all(csp->headers);
2212 * XXX: For consistency we might want to log a crunch message here.
2221 freez(csp->headers->first->str);
2222 build_request_line(csp, fwd, &csp->headers->first->str);
2225 hdr = list_to_text(csp->headers);
2228 /* FIXME Should handle error properly */
2229 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2233 * We have a request. Check if one of the crunchers wants it.
2235 if (crunch_response_triggered(csp, crunchers_all))
2238 * Yes. The client got the crunch response
2239 * and we are done here after cleaning up.
2242 list_remove_all(csp->headers);
2248 * The headers can't be removed earlier because
2249 * they were still needed for the referrer check
2250 * in case of CGI crunches.
2252 * XXX: Would it be worth to move the referrer check
2253 * into client_referrer() and set a flag if it's trusted?
2255 list_remove_all(csp->headers);
2257 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2259 if (fwd->forward_host)
2261 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2262 fwd->forward_host, fwd->forward_port, http->hostport);
2266 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2269 /* here we connect to the server, gateway, or the forwarder */
2271 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2272 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2274 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2275 forwarded_connect_retries, http->hostport);
2278 if (csp->sfd == JB_INVALID_SOCKET)
2280 if (fwd->type != SOCKS_NONE)
2283 rsp = error_response(csp, "forwarding-failed", errno);
2285 else if (errno == EINVAL)
2287 rsp = error_response(csp, "no-such-domain", errno);
2291 rsp = error_response(csp, "connect-failed", errno);
2292 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2297 /* Write the answer to the client */
2300 send_crunch_response(csp, rsp);
2307 if (fwd->forward_host || (http->ssl == 0))
2309 /* write the client's (modified) header to the server
2310 * (along with anything else that may be in the buffer)
2313 if (write_socket(csp->sfd, hdr, strlen(hdr))
2314 || (flush_socket(csp->sfd, csp) < 0))
2316 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2319 rsp = error_response(csp, "connect-failed", errno);
2323 send_crunch_response(csp, rsp);
2333 * We're running an SSL tunnel and we're not forwarding,
2334 * so just send the "connect succeeded" message to the
2335 * client, flush the rest, and get out of the way.
2337 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2345 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2347 /* we're finished with the client's header */
2350 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2352 /* pass data between the client and server
2353 * until one or the other shuts down the connection.
2362 * FD_ZERO here seems to point to an errant macro which crashes.
2363 * So do this by hand for now...
2365 memset(&rfds,0x00,sizeof(fd_set));
2369 FD_SET(csp->cfd, &rfds);
2370 FD_SET(csp->sfd, &rfds);
2372 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2376 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2380 /* this is the body of the browser's request
2381 * just read it and write it.
2384 if (FD_ISSET(csp->cfd, &rfds))
2386 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2390 break; /* "game over, man" */
2393 if (write_socket(csp->sfd, buf, (size_t)len))
2395 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2402 * The server wants to talk. It could be the header or the body.
2403 * If `hdr' is null, then it's the header otherwise it's the body.
2404 * FIXME: Does `hdr' really mean `host'? No.
2408 if (FD_ISSET(csp->sfd, &rfds))
2411 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2415 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2417 if (http->ssl && (fwd->forward_host == NULL))
2420 * Just hang up. We already confirmed the client's CONNECT
2421 * request with status code 200 and unencrypted content is
2422 * no longer welcome.
2424 log_error(LOG_LEVEL_ERROR,
2425 "CONNECT already confirmed. Unable to tell the client about the problem.");
2428 else if (byte_count)
2431 * Just hang up. We already transmitted the original headers
2432 * and parts of the original content and therefore missed the
2433 * chance to send an error message (without risking data corruption).
2435 * XXX: we could retry with a fancy range request here.
2437 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2438 "Unable to tell the client about the problem.");
2442 rsp = error_response(csp, "connect-failed", errno);
2446 send_crunch_response(csp, rsp);
2452 /* Add a trailing zero. This lets filter_popups
2453 * use string operations.
2457 #ifdef FEATURE_KILL_POPUPS
2458 /* Filter the popups on this read. */
2459 if (block_popups_now)
2461 filter_popups(buf, csp);
2463 #endif /* def FEATURE_KILL_POPUPS */
2465 /* Normally, this would indicate that we've read
2466 * as much as the server has sent us and we can
2467 * close the client connection. However, Microsoft
2468 * in its wisdom has released IIS/5 with a bug that
2469 * prevents it from sending the trailing \r\n in
2470 * a 302 redirect header (and possibly other headers).
2471 * To work around this if we've haven't parsed
2472 * a full header we'll append a trailing \r\n
2473 * and see if this now generates a valid one.
2475 * This hack shouldn't have any impacts. If we've
2476 * already transmitted the header or if this is a
2477 * SSL connection, then we won't bother with this
2478 * hack. So we only work on partially received
2479 * headers. If we append a \r\n and this still
2480 * doesn't generate a valid header, then we won't
2481 * transmit anything to the client.
2486 if (server_body || http->ssl)
2489 * If we have been buffering up the document,
2490 * now is the time to apply content modification
2491 * and send the result to the client.
2495 p = execute_content_filter(csp, content_filter);
2497 * If the content filter fails, use the original
2498 * buffer and length.
2499 * (see p != NULL ? p : csp->iob->cur below)
2503 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2506 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2508 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2511 hdr = list_to_text(csp->headers);
2514 /* FIXME Should handle error properly */
2515 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2519 * Shouldn't happen because this was the second sed run
2520 * and tags are only created for the first one.
2522 assert(!crunch_response_triggered(csp, crunchers_all));
2524 if (write_socket(csp->cfd, hdr, strlen(hdr))
2525 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2527 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2537 break; /* "game over, man" */
2541 * This is NOT the body, so
2542 * Let's pretend the server just sent us a blank line.
2544 log_error(LOG_LEVEL_INFO,
2545 "Malformerd HTTP headers detected and MS IIS5 hack enabled. "
2546 "Expect an invalid response or even no response at all.");
2547 snprintf(buf, sizeof(buf), "\r\n");
2548 len = (int)strlen(buf);
2551 * Now, let the normal header parsing algorithm below do its
2552 * job. If it fails, we'll exit instead of continuing.
2559 * If this is an SSL connection or we're in the body
2560 * of the server document, just write it to the client,
2561 * unless we need to buffer the body for later content-filtering
2564 if (server_body || http->ssl)
2569 * If there is no memory left for buffering the content, or the buffer limit
2570 * has been reached, switch to non-filtering mode, i.e. make & write the
2571 * header, flush the iob and buf, and get out of the way.
2573 if (add_to_iob(csp, buf, len))
2578 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2580 hdr = list_to_text(csp->headers);
2584 * Memory is too tight to even generate the header.
2585 * Send our static "Out-of-memory" page.
2587 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2588 rsp = cgi_error_memory();
2589 send_crunch_response(csp, rsp);
2593 hdrlen = strlen(hdr);
2595 if (write_socket(csp->cfd, hdr, hdrlen)
2596 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
2597 || (write_socket(csp->cfd, buf, (size_t)len)))
2599 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2606 * Reset the byte_count to the amount of bytes
2607 * we just flushed. len will be added a few lines below,
2608 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2610 byte_count = (size_t)flushed;
2612 content_filter = NULL;
2618 if (write_socket(csp->cfd, buf, (size_t)len))
2620 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2624 byte_count += (size_t)len;
2629 /* we're still looking for the end of the
2630 * server's header ... (does that make header
2631 * parsing an "out of body experience" ?
2635 * buffer up the data we just read. If that fails,
2636 * there's little we can do but send our static
2637 * out-of-memory page.
2639 if (add_to_iob(csp, buf, len))
2641 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2642 rsp = cgi_error_memory();
2643 send_crunch_response(csp, rsp);
2648 /* Convert iob into something sed() can digest */
2649 if (JB_ERR_PARSE == get_server_headers(csp))
2653 /* Well, we tried our MS IIS/5
2654 * hack and it didn't work.
2655 * The header is incomplete
2656 * and there isn't anything
2657 * we can do about it.
2663 /* Since we have to wait for
2664 * more from the server before
2665 * we can parse the headers
2666 * we just continue here.
2672 /* Did we actually get anything? */
2673 if (NULL == csp->headers->first)
2675 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2676 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2677 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2678 free_http_request(http);
2682 /* we have now received the entire header.
2683 * filter it and send the result to the client
2685 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2687 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2689 hdr = list_to_text(csp->headers);
2692 /* FIXME Should handle error properly */
2693 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2696 if (crunch_response_triggered(csp, crunchers_light))
2699 * One of the tags created by a server-header
2700 * tagger triggered a crunch. We already
2701 * delivered the crunch response to the client
2702 * and are done here after cleaning up.
2707 /* Buffer and pcrs filter this if appropriate. */
2709 if (!http->ssl) /* We talk plaintext */
2712 #ifdef FEATURE_KILL_POPUPS
2713 /* Start blocking popups if appropriate. */
2714 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2715 (csp->action->flags & ACTION_NO_POPUPS) != 0) /* Policy allows */
2717 block_popups_now = 1;
2719 * Filter the part of the body that came in the same read
2720 * as the last headers:
2722 filter_popups(csp->iob->cur, csp);
2724 #endif /* def FEATURE_KILL_POPUPS */
2725 content_filter = get_filter_function(csp);
2728 * Only write if we're not buffering for content modification
2730 if (!content_filter)
2732 /* write the server's (modified) header to
2733 * the client (along with anything else that
2734 * may be in the buffer)
2737 if (write_socket(csp->cfd, hdr, strlen(hdr))
2738 || ((len = flush_socket(csp->cfd, csp)) < 0))
2740 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2742 /* the write failed, so don't bother
2743 * mentioning it to the client...
2744 * it probably can't hear us anyway.
2750 byte_count += (size_t)len;
2753 /* we're finished with the server's header */
2758 /* If this was a MS IIS/5 hack then it means
2759 * the server has already closed the
2760 * connection. Nothing more to read. Time
2771 return; /* huh? we should never get here */
2774 if (csp->content_length == 0)
2777 * If Privoxy didn't recalculate the
2778 * Content-Lenght, byte_count is still
2781 csp->content_length = byte_count;
2784 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2785 csp->ip_addr_str, http->ocmd, csp->content_length);
2789 /*********************************************************************
2793 * Description : This is little more than chat. We only "serve" to
2794 * to close any socket that chat may have opened.
2797 * 1 : csp = Current client state (buffers, headers, etc...)
2801 *********************************************************************/
2803 void serve(struct client_state *csp)
2804 #else /* ifndef AMIGA */
2805 static void serve(struct client_state *csp)
2806 #endif /* def AMIGA */
2809 close_socket(csp->cfd);
2811 if (csp->sfd != JB_INVALID_SOCKET)
2813 close_socket(csp->sfd);
2816 csp->flags &= ~CSP_FLAG_ACTIVE;
2822 /*********************************************************************
2824 * Function : server_thread
2826 * Description : We only exist to call `serve' in a threaded environment.
2829 * 1 : data = Current client state (buffers, headers, etc...)
2831 * Returns : Always 0.
2833 *********************************************************************/
2834 static int32 server_thread(void *data)
2836 serve((struct client_state *) data);
2843 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2844 /*********************************************************************
2848 * Description : Print usage info & exit.
2850 * Parameters : Pointer to argv[0] for identifying ourselves
2854 *********************************************************************/
2855 static void usage(const char *myname)
2857 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2861 #endif /* defined(unix) */
2864 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2865 #endif /* defined(unix) */
2866 "[--version] [configfile]\n"
2867 "Aborting\n", myname);
2872 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2875 /*********************************************************************
2877 * Function : initialize_mutexes
2879 * Description : Prepares mutexes if mutex support is available.
2883 * Returns : Void, exits in case of errors.
2885 *********************************************************************/
2886 static void initialize_mutexes(void)
2890 #ifdef FEATURE_PTHREAD
2892 * Prepare global mutex semaphores
2894 err = pthread_mutex_init(&log_mutex, 0);
2896 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2899 * XXX: The assumptions below are a bit naive
2900 * and can cause locks that aren't necessary.
2902 * For example older FreeBSD versions (< 6.x?)
2903 * have no gethostbyname_r, but gethostbyname is
2906 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2907 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2908 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2910 * XXX: should we use a single mutex for
2911 * localtime() and gmtime() as well?
2913 #ifndef HAVE_GMTIME_R
2914 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2915 #endif /* ndef HAVE_GMTIME_R */
2917 #ifndef HAVE_LOCALTIME_R
2918 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2919 #endif /* ndef HAVE_GMTIME_R */
2922 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2923 #endif /* ndef HAVE_RANDOM */
2924 #endif /* FEATURE_PTHREAD */
2927 * TODO: mutex support for mingw32 would be swell.
2932 printf("Fatal error. Mutex initialization failed: %s.\n",
2941 /*********************************************************************
2945 * Description : Load the config file and start the listen loop.
2946 * This function is a lot more *sane* with the `load_config'
2947 * and `listen_loop' functions; although it stills does
2948 * a *little* too much for my taste.
2951 * 1 : argc = Number of parameters (including $0).
2952 * 2 : argv = Array of (char *)'s to the parameters.
2954 * Returns : 1 if : can't open config file, unrecognized directive,
2955 * stats requested in multi-thread mode, can't open the
2956 * log file, can't open the jar file, listen port is invalid,
2957 * any load fails, and can't bind port.
2959 * Else main never returns, the process must be signaled
2960 * to terminate execution. Or, on Windows, use the
2961 * "File", "Exit" menu option.
2963 *********************************************************************/
2965 int real_main(int argc, const char *argv[])
2967 int main(int argc, const char *argv[])
2971 unsigned int random_seed;
2973 struct passwd *pw = NULL;
2974 struct group *grp = NULL;
2977 char *pre_chroot_nslookup_to_load_resolver = NULL;
2984 #if !defined(_WIN32)
2992 * Parse the command line arguments
2994 * XXX: simply printing usage information in case of
2995 * invalid arguments isn't particular user friendly.
2997 while (++argc_pos < argc)
3000 /* Check to see if the service must be installed or uninstalled */
3001 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3003 const char *pName = argv[argc_pos] + 9;
3006 exit( (install_service(pName)) ? 0 : 1 );
3008 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3010 const char *pName = argv[argc_pos] + 11;
3013 exit((uninstall_service(pName)) ? 0 : 1);
3015 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3017 bRunAsService = TRUE;
3018 w32_set_service_cwd();
3019 atexit(w32_service_exit_notify);
3022 #endif /* defined(_WIN32) */
3025 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3027 if (strcmp(argv[argc_pos], "--help") == 0)
3032 else if(strcmp(argv[argc_pos], "--version") == 0)
3034 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3040 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3045 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3047 if (++argc_pos == argc) usage(argv[0]);
3048 pidfile = strdup(argv[argc_pos]);
3051 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3053 if (++argc_pos == argc) usage(argv[argc_pos]);
3055 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3058 if (NULL == (grp = getgrnam(p)))
3060 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3064 if (NULL == (pw = getpwnam(argv[argc_pos])))
3066 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3069 if (p != NULL) *--p = '\0';
3072 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3074 if (++argc_pos == argc) usage(argv[0]);
3075 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3078 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3082 #endif /* defined(unix) */
3084 else if (argc_pos + 1 != argc)
3087 * This is neither the last command line
3088 * option, nor was it recognized before,
3089 * therefore it must be invalid.
3095 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3097 configfile = argv[argc_pos];
3100 } /* -END- while (more arguments) */
3103 if ( *configfile != '/' )
3105 char cwd[BUFFER_SIZE];
3107 size_t abs_file_size;
3109 /* make config-filename absolute here */
3110 if (NULL == getcwd(cwd, sizeof(cwd)))
3112 perror("failed to get current working directory");
3117 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3118 basedir = strdup(cwd);
3120 if (NULL == basedir ||
3121 NULL == (abs_file = malloc(abs_file_size)))
3123 perror("malloc failed");
3126 strlcpy(abs_file, basedir, abs_file_size);
3127 strlcat(abs_file, "/", abs_file_size );
3128 strlcat(abs_file, configfile, abs_file_size);
3129 configfile = abs_file;
3131 #endif /* defined unix */
3135 clients->next = NULL;
3137 /* XXX: factor out initialising after the next stable release. */
3140 #elif defined(_WIN32)
3144 /* Prepare mutexes if supported and necessary. */
3145 initialize_mutexes();
3147 /* Enable logging until further notice. */
3148 init_log_module(Argv[0]);
3150 random_seed = (unsigned int)time(NULL);
3152 srandom(random_seed);
3153 #elif defined (_WIN32)
3155 * See pick_from_range() in miscutil.c for details.
3157 log_error(LOG_LEVEL_INFO,
3158 "No thread-safe PRNG implemented for your platform. "
3159 "Using weak \'randomization\' factor which will "
3160 "limit the already questionable usefulness of "
3161 "header-time-randomizing actions (disabled by default).");
3164 #endif /* ifdef HAVE_RANDOM */
3167 * Unix signal handling
3169 * Catch the abort, interrupt and terminate signals for a graceful exit
3170 * Catch the hangup signal so the errlog can be reopened.
3171 * Ignore the broken pipe signals (FIXME: Why?)
3173 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3176 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3177 const int ignored_signals[] = { SIGPIPE, 0 };
3179 for (idx = 0; catched_signals[idx] != 0; idx++)
3181 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3182 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3184 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3185 #endif /* ifdef sun */
3187 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3191 for (idx = 0; ignored_signals[idx] != 0; idx++)
3193 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3195 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3200 #else /* ifdef _WIN32 */
3201 # ifdef _WIN_CONSOLE
3203 * We *are* in a windows console app.
3204 * Print a verbose messages about FAQ's and such
3206 printf("%s", win32_blurb);
3207 # endif /* def _WIN_CONSOLE */
3208 #endif /* def _WIN32 */
3211 /* Initialize the CGI subsystem */
3212 cgi_init_error_messages();
3215 * If runnig on unix and without the --nodaemon
3216 * option, become a daemon. I.e. fork, detach
3217 * from tty and get process group leadership
3230 if ( pid < 0 ) /* error */
3235 else if ( pid != 0 ) /* parent */
3240 * must check for errors
3241 * child died due to missing files aso
3244 wpid = waitpid( pid, &status, WNOHANG );
3253 /* Should be more portable, but not as well tested */
3258 #else /* ndef __FreeBSD__ */
3260 #endif /* ndef __FreeBSD__ */
3261 fd = open("/dev/tty", O_RDONLY);
3264 /* no error check here */
3265 ioctl( fd, TIOCNOTTY,0 );
3270 * stderr (fd 2) will be closed later on, when the
3271 * log file has been parsed.
3278 } /* -END- if (!no_daemon) */
3281 * As soon as we have written the PID file, we can switch
3282 * to the user and group ID indicated by the --user option
3288 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3290 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3296 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3298 /* Read the time zone file from /etc before doing chroot. */
3300 if (NULL != pre_chroot_nslookup_to_load_resolver
3301 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3303 /* Initialize resolver library. */
3304 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3306 if (chroot(pw->pw_dir) < 0)
3308 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3312 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3315 if (setuid(pw->pw_uid))
3317 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3321 char putenv_dummy[64];
3323 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3324 if (putenv(putenv_dummy) != 0)
3326 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3329 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3330 if (putenv(putenv_dummy) != 0)
3332 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3338 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3341 #endif /* defined unix */
3344 /* This will be FALSE unless the command line specified --service
3348 /* Yup, so now we must attempt to establish a connection
3349 * with the service dispatcher. This will only work if this
3350 * process was launched by the service control manager to
3351 * actually run as a service. If this isn't the case, i've
3352 * known it take around 30 seconds or so for the call to return.
3355 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3356 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3358 /* Service has run, and at this point is now being stopped, so just return */
3363 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3365 /* An error occurred. Usually it's because --service was wrongly specified
3366 * and we were unable to connect to the Service Control Dispatcher because
3367 * it wasn't expecting us and is therefore not listening.
3369 * For now, just continue below to call the listen_loop function.
3372 #endif /* def _WIN32 */
3382 /*********************************************************************
3384 * Function : bind_port_helper
3386 * Description : Bind the listen port. Handles logging, and aborts
3390 * 1 : config = Privoxy configuration. Specifies port
3393 * Returns : Port that was opened.
3395 *********************************************************************/
3396 static jb_socket bind_port_helper(struct configuration_spec * config)
3401 if ( (config->haddr != NULL)
3402 && (config->haddr[0] == '1')
3403 && (config->haddr[1] == '2')
3404 && (config->haddr[2] == '7')
3405 && (config->haddr[3] == '.') )
3407 log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only",
3410 else if (config->haddr == NULL)
3412 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3417 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3418 config->hport, config->haddr);
3421 result = bind_port(config->haddr, config->hport, &bfd);
3428 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3429 "There may be another Privoxy or some other "
3430 "proxy running on port %d",
3431 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3432 config->hport, config->hport);
3435 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3436 "The hostname is not resolvable",
3437 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3440 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3441 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3444 /* shouldn't get here */
3445 return JB_INVALID_SOCKET;
3448 config->need_bind = 0;
3455 /* Without this simple workaround we get this compiler warning from _beginthread
3456 * warning C4028: formal parameter 1 different from declaration
3458 void w32_service_listen_loop(void *p)
3462 #endif /* def _WIN32 */
3465 /*********************************************************************
3467 * Function : listen_loop
3469 * Description : bind the listen port and enter a "FOREVER" listening loop.
3475 *********************************************************************/
3476 static void listen_loop(void)
3478 struct client_state *csp = NULL;
3480 struct configuration_spec * config;
3482 config = load_config();
3484 bfd = bind_port_helper(config);
3486 #ifdef FEATURE_GRACEFUL_TERMINATION
3487 while (!g_terminate)
3492 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3493 while (waitpid(-1, NULL, WNOHANG) > 0)
3495 /* zombie children */
3497 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3500 * Free data that was used by died threads
3506 * Re-open the errlog after HUP signal
3508 if (received_hup_signal)
3510 init_error_log(Argv[0], config->logfile);
3511 received_hup_signal = 0;
3516 #ifdef FEATURE_COOKIE_JAR
3518 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3519 * already open. I think unload_configfile() is not being run, which should do
3520 * this work. Until that can get resolved, we'll use this workaround.
3524 if (csp->config->jar)
3526 fclose(csp->config->jar);
3527 csp->config->jar = NULL;
3529 #endif /* FEATURE_COOKIE_JAR */
3530 #endif /* __OS2__ */
3532 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3534 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3538 csp->flags |= CSP_FLAG_ACTIVE;
3539 csp->sfd = JB_INVALID_SOCKET;
3541 csp->config = config = load_config();
3543 if ( config->need_bind )
3546 * Since we were listening to the "old port", we will not see
3547 * a "listen" param change until the next IJB request. So, at
3548 * least 1 more request must be made for us to find the new
3549 * setting. I am simply closing the old socket and binding the
3552 * Which-ever is correct, we will serve 1 more page via the
3553 * old settings. This should probably be a "show-proxy-args"
3554 * request. This should not be a so common of an operation
3555 * that this will hurt people's feelings.
3560 bfd = bind_port_helper(config);
3563 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3565 if (!accept_connection(csp, bfd))
3567 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3580 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3583 #ifdef FEATURE_TOGGLE
3584 if (global_toggle_state)
3585 #endif /* def FEATURE_TOGGLE */
3587 csp->flags |= CSP_FLAG_TOGGLED_ON;
3590 if (run_loader(csp))
3592 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3593 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3597 if (block_acl(NULL,csp))
3599 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3600 close_socket(csp->cfd);
3604 #endif /* def FEATURE_ACL */
3606 /* add it to the list of clients */
3607 csp->next = clients->next;
3608 clients->next = csp;
3610 if (config->multi_threaded)
3614 /* this is a switch () statment in the C preprocessor - ugh */
3615 #undef SELECTED_ONE_OPTION
3617 /* Use Pthreads in preference to native code */
3618 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3619 #define SELECTED_ONE_OPTION
3621 pthread_t the_thread;
3622 pthread_attr_t attrs;
3624 pthread_attr_init(&attrs);
3625 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3626 errno = pthread_create(&the_thread, &attrs,
3627 (void * (*)(void *))serve, csp);
3628 child_id = errno ? -1 : 0;
3629 pthread_attr_destroy(&attrs);
3633 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3634 #define SELECTED_ONE_OPTION
3635 child_id = _beginthread(
3636 (void (*)(void *))serve,
3641 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3642 #define SELECTED_ONE_OPTION
3643 child_id = _beginthread(
3644 (void(* _Optlink)(void*))serve,
3650 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3651 #define SELECTED_ONE_OPTION
3653 thread_id tid = spawn_thread
3654 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3656 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3658 child_id = (int) tid;
3667 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3668 #define SELECTED_ONE_OPTION
3669 csp->cfd = ReleaseSocket(csp->cfd, -1);
3672 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3673 NP_Output, Output(),
3674 NP_CloseOutput, FALSE,
3675 NP_Name, (ULONG)"privoxy child",
3679 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3680 NP_Output, Output(),
3681 NP_CloseOutput, FALSE,
3682 NP_Name, (ULONG)"privoxy child",
3683 NP_StackSize, 200*1024,
3689 ((struct Task *)child_id)->tc_UserData = csp;
3690 Signal((struct Task *)child_id, SIGF_SINGLE);
3695 #if !defined(SELECTED_ONE_OPTION)
3698 /* This block is only needed when using fork().
3699 * When using threads, the server thread was
3700 * created and run by the call to _beginthread().
3702 if (child_id == 0) /* child */
3705 #ifdef FEATURE_TOGGLE
3706 int inherited_toggle_state = global_toggle_state;
3707 #endif /* def FEATURE_TOGGLE */
3712 * If we've been toggled or we've blocked the request, tell Mom
3715 #ifdef FEATURE_TOGGLE
3716 if (inherited_toggle_state != global_toggle_state)
3718 rc |= RC_FLAG_TOGGLED;
3720 #endif /* def FEATURE_TOGGLE */
3722 #ifdef FEATURE_STATISTICS
3723 if (csp->flags & CSP_FLAG_REJECTED)
3725 rc |= RC_FLAG_BLOCKED;
3727 #endif /* ndef FEATURE_STATISTICS */
3731 else if (child_id > 0) /* parent */
3733 /* in a fork()'d environment, the parent's
3734 * copy of the client socket and the CSP
3738 #if !defined(_WIN32) && !defined(__CYGWIN__)
3740 wait( &child_status );
3743 * Evaluate child's return code: If the child has
3744 * - been toggled, toggle ourselves
3745 * - blocked its request, bump up the stats counter
3748 #ifdef FEATURE_TOGGLE
3749 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3751 global_toggle_state = !global_toggle_state;
3753 #endif /* def FEATURE_TOGGLE */
3755 #ifdef FEATURE_STATISTICS
3757 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3761 #endif /* def FEATURE_STATISTICS */
3763 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3764 close_socket(csp->cfd);
3765 csp->flags &= ~CSP_FLAG_ACTIVE;
3769 #undef SELECTED_ONE_OPTION
3770 /* end of cpp switch () */
3772 if (child_id < 0) /* failed */
3774 char buf[BUFFER_SIZE];
3776 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3778 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3780 write_socket(csp->cfd, buf, strlen(buf));
3781 close_socket(csp->cfd);
3782 csp->flags &= ~CSP_FLAG_ACTIVE;
3793 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3795 /* Clean up. Aim: free all memory (no leaks) */
3796 #ifdef FEATURE_GRACEFUL_TERMINATION
3798 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3800 unload_current_config_file();
3801 unload_current_actions_file();
3802 unload_current_re_filterfile();
3803 #ifdef FEATURE_TRUST
3804 unload_current_trust_file();
3807 if (config->multi_threaded)
3814 } while ((clients->next != NULL) && (--i > 0));
3818 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3829 #ifdef FEATURE_COOKIE_JAR
3830 if (NULL != config->jar)
3832 fclose(config->jar);
3836 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3837 /* Cleanup - remove taskbar icon etc. */
3842 #endif /* FEATURE_GRACEFUL_TERMINATION */