1 const char jcc_rcs[] = "$Id: jcc.c,v 1.151 2007/09/29 10:21:16 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 1.151 2007/09/29 10:21:16 fabiankeil
37 * - Move get_filter_function() from jcc.c to filters.c
38 * so the filter functions can be static.
39 * - Don't bother filtering body-less responses.
41 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
42 * Execute content filters through execute_content_filter().
44 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
45 * Initialize req to NULL to make sure it's defined if the
46 * first read_socket() call fails. Reported by icmp30.
48 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
49 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
50 * extensive comments moved to user manual.
52 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
53 * Don't crash if a broken header filter wiped out the request line.
55 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
56 * Fix byte_count calculation in case of flushes
57 * and don't parse the server headers a second time.
59 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
60 * - If there's a connection problem after we already forwarded
61 * parts of the original content, just hang up. Fixes BR#1776724.
62 * - Fix warnings about unused code on mingw32.
63 * - In case of flushes, calculate the byte count
64 * less incorrectly (I think).
66 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
67 * Add some more prototypes for static functions.
69 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
70 * Comment out request_contains_null_bytes() until it's used again.
72 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
73 * #1763173 from Stefan Huehner: s@const static@static const@
74 * and declare some more functions static.
76 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
77 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
78 * and explain why they were rejected in the first place.
79 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
80 * CONNECT requests. The request line was missing.
81 * - Add two more XXX reminders as we don't have enough already.
83 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
84 * As Hal noticed, checking dispatch_cgi() as the last cruncher
85 * looks like a bug if CGI requests are blocked unintentionally,
86 * so don't do it unless the user enabled the new config option
87 * "allow-cgi-request-crunching".
89 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
90 * - Allow to rewrite the request destination behind the client's back.
91 * - Turn the weird-looking unconditional for loop that
92 * reads the client request into a conditional while loop.
93 * Move the stuff that only runs once out of the loop.
94 * - Move parts of chat(), server_content_type() and the
95 * necessary stuff to fix BR#1750917 into get_filter_function().
97 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
98 * Temporary workaround for BR#1730105.
100 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
101 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
102 * deadlocks and crashes on OpenBSD and possibly other OS with neither
103 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
104 * Thanks to Ralf Horstmann for report and solution.
106 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
107 * Add forward-override{} to change the forwarding settings through
108 * action sections. This is mainly interesting to forward different
109 * clients differently (for example based on User-Agent or request
112 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
113 * - Let usage() mention the --chroot parameter.
114 * - Use read_socket() consistently and always leave
115 * the last buffer byte alone, even in cases where
116 * null termination (currently) doesn't matter.
118 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
119 * - Fix config file loading on Unix if no config file is specified.
120 * Since r1.97 Privoxy would always interpret the last argument as
121 * config file, even if it's a valid command line option.
122 * - Abort in case of unrecognized command line options. Closes #1719696.
123 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
124 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
126 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
127 * - Don't rerun crunchers that only depend on the request URL.
128 * - Don't count redirects and CGI requests as "blocked requests".
130 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
131 * Support crunching based on tags created by server-header taggers.
133 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
134 * Make HTTP snippets static (again). Add a Content-Type for those
135 * with content so the browser doesn't guess it based on the URL.
137 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
138 * Move crunching and request line rebuilding out of chat().
140 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
141 * Introduce tags as alternative way to specify which
142 * actions apply to a request. At the moment tags can be
143 * created based on client and server headers.
145 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
146 * Don't CLF-log CONNECT requests twice.
148 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
149 * Log the source address for ACL-related connection drops.
151 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
152 * New config option: enforce-blocks.
154 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
155 * - Move null byte check into separate function.
156 * - Don't confuse the client with error pages
157 * if a CONNECT request was already confirmed.
159 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
160 * Speed up NULL byte escaping and only log the complete
161 * NULL byte requests with header debugging enabled.
163 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
164 * Answer requests that contain NULL bytes with
165 * a custom response instead of waiting for more
166 * data until the client eventually hangs up.
168 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
169 * - Move delivery and logging of crunched responses
170 * from chat() into send_crunch_response().
171 * - Display the reason for generating http_responses.
172 * - Log the content length for LOG_LEVEL_CLF correctly
173 * (still incorrect for some fixed responses).
174 * - Reword an incorrect comment about
175 * treat-forbidden-connects-like-blocks violating
177 * - Add some log messages.
179 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
180 * Move mutex initialization into separate
181 * function and exit in case of errors.
183 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
184 * - Start to reduce chat()'s line count and move
185 * parts of it into separate functions.
186 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
188 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
189 * - Add Proxy-Agent header to HTTP snippets that are
190 * supposed to reach HTTP clients only.
191 * - Made a few CONNECT log messages more descriptive.
192 * - Catch completely empty server responses (as seen
193 * with Tor's fake ".noconnect" top level domain).
194 * - Use shiny new "forwarding-failed" template for socks errors.
196 * Revision 1.118 2007/01/07 07:43:43 joergs
197 * AmigaOS4 support added.
199 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
200 * Added config option accept-intercepted-requests
201 * and disabled it by default.
203 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
204 * Reverted parts of my last commit
205 * to keep error handling working.
207 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
208 * Fixed gcc43 conversion warnings.
210 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
211 * Fix -pedantic ISO C warning about converting
212 * from function pointer to object pointer.
214 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
215 * Silence compiler warning I introduced with my last commit.
217 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
218 * Mutex protect rand() if POSIX threading
219 * is used, warn the user if that's not possible
220 * and stop using it on _WIN32 where it could
223 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
224 * Don't prevent core dumps by catching SIGABRT.
225 * It's rude and makes debugging unreasonable painful.
227 * Revision 1.110 2006/12/13 14:52:53 etresoft
228 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
230 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
231 * Privoxy is now able to run as intercepting
232 * proxy in combination with any packet filter
233 * that does the port redirection. The destination
234 * is extracted from the "Host:" header which
235 * should be available for nearly all requests.
237 * Moved HTTP snipplets into jcc.c.
238 * Added error message for gopher proxy requests.
240 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
241 * Only unlink the pidfile if it's actually used.
243 * Change order of interception checks to make
244 * it possible to block or redirect requests for
247 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
248 * Make pthread mutex locking more generic. Instead of
249 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
250 * and use mutex locking unless there is an _r function
251 * available. Better safe than sorry.
253 * Fixes "./configure --disable-pthread" and should result
254 * in less threading-related problems on pthread-using platforms,
255 * but it still doesn't fix BR#1122404.
257 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
258 * Move pthread.h inclusion from jcc.c to jcc.h.
259 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
261 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
262 * Don't exit after receiving the second SIGHUP on Solaris.
264 * Fixes BR 1052235, but the same problem may exist on other
265 * systems. Once 3.0.6 is out we should use sigset()
266 * where available and see if it breaks anything.
268 * Revision 1.104 2006/09/23 13:26:38 roro
269 * Replace TABs by spaces in source code.
271 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
272 * Fix +redirect{}. Didn't work with -fast-redirects.
274 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
275 * Respond with 400 and a short text message
276 * if the client tries to use Privoxy as FTP proxy.
278 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
279 * Make number of retries in case of forwarded-connect problems
280 * a config file option (forwarded-connect-retries) and use 0 as
283 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
284 * Set random(3) seed.
286 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
287 * Follow the OpenBSD port's lead and protect the resolve
288 * functions on OpenBSD as well.
290 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
291 * --user fix. Only use the user as group if no group is specified.
292 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
294 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
295 * Windows service (re-)integration
299 * --install[:service_name]
300 * --uninstall[:service_name]
303 * They work as follows:
304 * --install will create a service for you and then terminate.
305 * By default the service name will be "privoxy" (without the quotes).
306 * However you can run multiple services if you wish, just by adding
307 * a colon and then a name (no spaces).
309 * --uninstall follows the exact same rules a --install.
311 * --service is used when the program is executed by the service
312 * control manager, and in normal circumstances would never be
313 * used as a command line argument.
315 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
316 * Windows service integration
318 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
319 * Incorporate Fabian Keil's patch work:
320 http://www.fabiankeil.de/sourcecode/privoxy/
322 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
323 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
324 * with what was really the latest development (the v_3_0_branch branch)
326 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
327 * Thanks to Jindrich Makovicka for a race condition fix for the log
328 * file. The race condition remains for non-pthread implementations.
329 * Reference patch #1175720.
331 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
332 * Add the ability to check jpeg images for invalid
333 * lengths of comment blocks. Defensive strategy
334 * against the exploit:
335 * Microsoft Security Bulletin MS04-028
336 * Buffer Overrun in JPEG Processing (GDI+) Could
337 * Allow Code Execution (833987)
338 * Enabled with +inspect-jpegs in actions files.
340 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
341 * - Fixed usage info for non-unix platforms
342 * - Fixed small cmdline parsing bug
344 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
345 * Diagnostics: Now preserve the returncode of pthread_create
346 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
348 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
349 * No longer ignore SIGCHLD. Fixes bug #769381
351 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
352 * Close jarfile on graceful exit, remove stray line
354 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
355 * Cosmetics: Killed a warning, a typo and an allocation left at exit
357 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
358 * No longer rely on non-POSIX.1 extensions of getcwd().
361 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
362 * Replaced setenv() by posix-compliant putenv()
363 * Thanks to Neil McCalden (nmcc AT users.sf.net).
365 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
366 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
368 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
369 * Clean-up and extension of improvements for forked mode:
370 * - Child's return code now consists of flags RC_FLAG_*
371 * - Reporting toggle to parent now properly #ifdef'ed
372 * - Children now report blocking to parent. This enables
373 * statistics in forked mode
375 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
376 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
377 * Children now report having been toggled through _exit(17), parents
378 * watch for that code and toggle themselves if found.
380 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
381 * Wrapping all *_r functions (the non-_r versions of them) with
382 * mutex semaphores for OSX. Hopefully this will take care of all
383 * of those pesky crash reports.
385 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
386 * Fixed two mostly harmless mem leaks
388 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
389 * Fix: Head of global clients list now initialized to NULL
391 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
392 * Added basic support for OPTIONS and TRACE HTTP methods:
393 * - New interceptor direct_response() added in chat().
394 * - sed() moved to earlier in the process, so that the
395 * Host: header is evaluated before actions and forwarding
398 * Revision 1.92 2002/05/08 16:00:46 oes
399 * Chat's buffer handling:
400 * - Fixed bug with unchecked out-of-mem conditions
401 * while reading client request & server headers
402 * - No longer predict if the buffer limit will be exceeded
403 * in the next read -- check add_to_iob's new
404 * return code. If buffer couldn't be extended
405 * (policy or out-of-mem) while
406 * - reading from client: abort
407 * - reading server headers: send error page
408 * - buffering server body for filter: flush,
409 * and if that fails: send error page
411 * Revision 1.91 2002/04/08 20:35:58 swa
414 * Revision 1.90 2002/04/02 14:57:28 oes
415 * Made sending wafers independent of FEATURE_COOKIE_JAR
417 * Revision 1.89 2002/03/31 17:18:59 jongfoster
418 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
420 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
421 * More compiler warning message maintenance
423 * Revision 1.87 2002/03/26 22:29:54 swa
424 * we have a new homepage!
426 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
427 * Workaround for closing the jarfile before load_config() comes around again
429 * Revision 1.85 2002/03/24 15:23:33 jongfoster
432 * Revision 1.84 2002/03/24 13:25:43 swa
433 * name change related issues
435 * Revision 1.83 2002/03/16 23:54:06 jongfoster
436 * Adding graceful termination feature, to help look for memory leaks.
437 * If you enable this (which, by design, has to be done by hand
438 * editing config.h) and then go to http://i.j.b/die, then the program
439 * will exit cleanly after the *next* request. It should free all the
440 * memory that was used.
442 * Revision 1.82 2002/03/13 00:27:05 jongfoster
445 * Revision 1.81 2002/03/12 01:42:50 oes
446 * Introduced modular filters
448 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
449 * OS/2 port maintenance:
450 * - Fixed EMX build - it had decayed a little
451 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
452 * substituted a memset for now.
454 * Revision 1.79 2002/03/09 20:03:52 jongfoster
455 * - Making various functions return int rather than size_t.
456 * (Undoing a recent change). Since size_t is unsigned on
457 * Windows, functions like read_socket that return -1 on
458 * error cannot return a size_t.
460 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
461 * crashes, and also frequently caused JB to jump to 100%
462 * CPU and stay there. (Because it thought it had just
463 * read ((unsigned)-1) == 4Gb of data...)
465 * - The signature of write_socket has changed, it now simply
466 * returns success=0/failure=nonzero.
468 * - Trying to get rid of a few warnings --with-debug on
469 * Windows, I've introduced a new type "jb_socket". This is
470 * used for the socket file descriptors. On Windows, this
471 * is SOCKET (a typedef for unsigned). Everywhere else, it's
472 * an int. The error value can't be -1 any more, so it's
473 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
474 * Windows it maps to the #define INVALID_SOCKET.)
476 * - The signature of bind_port has changed.
478 * Revision 1.78 2002/03/08 21:35:04 oes
479 * Added optional group supplement to --user option. Will now use default group of user if no group given
481 * Revision 1.77 2002/03/07 03:52:06 oes
482 * - Fixed compiler warnings etc
483 * - Improved handling of failed DNS lookups
485 * Revision 1.76 2002/03/06 22:54:35 jongfoster
486 * Automated function-comment nitpicking.
488 * Revision 1.75 2002/03/06 10:02:19 oes
489 * Fixed stupid bug when --user was not given
491 * Revision 1.74 2002/03/06 00:49:31 jongfoster
492 * Fixing warning on Windows
493 * Making #ifdefs that refer to the same variable consistently
494 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
496 * Revision 1.73 2002/03/05 23:57:30 hal9
497 * Stray character 's' on line 1618 was breaking build.
499 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
500 * - Re-enable OS/2 building after new parms were added
501 * - Fix false out of memory report when resolving CGI templates when no IP
502 * address is available of failed attempt (a la no such domain)
504 * Revision 1.71 2002/03/05 18:13:56 oes
505 * Added --user option
507 * Revision 1.70 2002/03/05 04:52:42 oes
508 * Deleted non-errlog debugging code
510 * Revision 1.69 2002/03/04 23:50:00 jongfoster
511 * Splitting off bind_port() call into bind_port_helper(), with
514 * Revision 1.68 2002/03/04 20:17:32 oes
517 * Revision 1.67 2002/03/04 18:18:57 oes
518 * - Removed _DEBUG mode
519 * - Cleand up cmdline parsing
520 * - Introduced --no-daemon, --pidfile options
521 * - Cleaned up signal handling:
522 * - Terminate cleanly on INT, TERM and ABRT
523 * - Schedule logfile for re-opening on HUP
524 * - Ignore CHLD and PIPE
525 * - Leave the rest with their default handlers
526 * - Uniform handler registration
527 * - Added usage() function
528 * - Played styleguide police
530 * Revision 1.66 2002/03/03 15:06:55 oes
531 * Re-enabled automatic config reloading
533 * Revision 1.65 2002/03/03 14:49:11 oes
534 * Fixed CLF logging: Now uses client's original HTTP request
536 * Revision 1.64 2002/03/03 09:18:03 joergs
537 * Made jumbjuster work on AmigaOS again.
539 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
540 * Clean up a little CRLF unpleasantness that suddenly appeared
542 * Revision 1.62 2002/02/20 23:17:23 jongfoster
543 * Detecting some out-of memory conditions and exiting with a log message.
545 * Revision 1.61 2002/01/17 21:01:52 jongfoster
546 * Moving all our URL and URL pattern parsing code to urlmatch.c.
548 * Revision 1.60 2001/12/30 14:07:32 steudten
549 * - Add signal handling (unix)
550 * - Add SIGHUP handler (unix)
551 * - Add creation of pidfile (unix)
552 * - Add action 'top' in rc file (RH)
553 * - Add entry 'SIGNALS' to manpage
554 * - Add exit message to logfile (unix)
556 * Revision 1.59 2001/12/13 14:07:18 oes
557 * Fixed Bug: 503 error page now sent OK
559 * Revision 1.58 2001/11/30 23:37:24 jongfoster
560 * Renaming the Win32 config file to config.txt - this is almost the
561 * same as the corresponding UNIX name "config"
563 * Revision 1.57 2001/11/16 00:47:43 jongfoster
564 * Changing the tty-disconnection code to use setsid().
566 * Revision 1.56 2001/11/13 20:20:54 jongfoster
567 * Tabs->spaces, fixing a bug with missing {} around an if()
569 * Revision 1.55 2001/11/13 20:14:53 jongfoster
570 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
572 * Revision 1.54 2001/11/07 00:03:14 steudten
573 * Give reliable return value if an error
574 * occurs not just 0 with new daemon mode.
576 * Revision 1.53 2001/11/05 21:41:43 steudten
577 * Add changes to be a real daemon just for unix os.
578 * (change cwd to /, detach from controlling tty, set
579 * process group and session leader to the own process.
581 * Add some fatal-error log message for failed malloc().
582 * Add '-d' if compiled with 'configure --with-debug' to
583 * enable debug output.
585 * Revision 1.52 2001/10/26 20:11:20 jongfoster
586 * Fixing type mismatch
588 * Revision 1.51 2001/10/26 17:38:28 oes
591 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
592 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
593 * threads to call select() simultaneously. So, it's time to do a real, live,
594 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
595 * (native). Both versions will work, but using __OS2__ offers multi-threading.
597 * Revision 1.49 2001/10/23 21:41:35 jongfoster
598 * Added call to initialize the (statically-allocated of course)
599 * "out of memory" CGI response.
601 * Revision 1.48 2001/10/10 19:56:46 jongfoster
602 * Moving some code that wasn't cookie-related out of an #ifdef
605 * Revision 1.47 2001/10/10 16:44:36 oes
606 * Added CONNECT destination port limitation check
608 * Revision 1.46 2001/10/08 15:17:41 oes
609 * Re-enabled SSL forwarding
611 * Revision 1.45 2001/10/07 15:42:11 oes
612 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
614 * Moved downgrading of the HTTP version from parse_http_request to
615 * chat(), since we can't decide if it is necessary before we have
616 * determined the actions for the URL. The HTTP command is now
617 * *always* re-built so the repairs need no longer be special-cased.
619 * filter_popups now gets a csp pointer so it can raise the new
620 * CSP_FLAG_MODIFIED flag.
624 * Added configurable size limit for the IOB. If the IOB grows so
625 * large that the next read would exceed the limit, the header
626 * is generated, and the header & unfiltered buffer are flushed
627 * to the client. Chat then continues in non-buffering,
628 * non-filtering body mode.
630 * Revision 1.44 2001/10/02 18:13:57 oes
633 * Revision 1.43 2001/10/02 15:32:13 oes
634 * Moved generation of hdr
636 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
637 * Cleaning up 2 compiler warnings on OS/2.
639 * Revision 1.41 2001/09/16 17:05:14 jongfoster
640 * Removing unused #include showarg.h
642 * Revision 1.40 2001/09/16 15:41:45 jongfoster
643 * Fixing signed/unsigned comparison warning.
645 * Revision 1.39 2001/09/16 13:21:27 jongfoster
646 * Changes to use new list functions.
648 * Revision 1.38 2001/09/16 13:01:46 jongfoster
649 * Removing redundant function call that zeroed zalloc()'d memory.
651 * Revision 1.37 2001/09/10 11:12:24 oes
652 * Deleted unused variable
654 * Revision 1.36 2001/09/10 10:56:15 oes
655 * Silenced compiler warnings
657 * Revision 1.35 2001/07/31 14:44:22 oes
658 * Deleted unused size parameter from filter_popups()
660 * Revision 1.34 2001/07/30 22:08:36 jongfoster
661 * Tidying up #defines:
662 * - All feature #defines are now of the form FEATURE_xxx
663 * - Permanently turned off WIN_GUI_EDIT
664 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
666 * Revision 1.33 2001/07/29 19:32:00 jongfoster
667 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
669 * Revision 1.32 2001/07/29 18:47:05 jongfoster
670 * Adding missing #include "loadcfg.h"
672 * Revision 1.31 2001/07/29 12:17:48 oes
673 * Applied pthread fix by Paul Lieverse
675 * Revision 1.30 2001/07/25 22:57:13 jongfoster
676 * __BEOS__ no longer overrides FEATURE_PTHREAD.
677 * This is because FEATURE_PTHREAD will soon be widely used, so I
678 * want to keep it simple.
680 * Revision 1.29 2001/07/24 12:47:06 oes
681 * Applied BeOS support update by Eugenia
683 * Revision 1.28 2001/07/23 13:26:12 oes
684 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
686 * Revision 1.27 2001/07/19 19:09:47 haroon
687 * - Added code to take care of the situation where while processing the first
688 * server response (which includes the server header), after finding the end
689 * of the headers we were not looking past the end of the headers for
690 * content modification. I enabled it for filter_popups.
691 * Someone else should look to see if other similar operations should be
692 * done to the discarded portion of the buffer.
694 * Note 2001/07/20: No, the other content modification mechanisms will process
695 * the whole iob later anyway. --oes
697 * Revision 1.26 2001/07/18 12:31:36 oes
700 * Revision 1.25 2001/07/15 19:43:49 jongfoster
701 * Supports POSIX threads.
702 * Also removed some unused #includes.
704 * Revision 1.24 2001/07/13 14:00:40 oes
705 * - Generic content modification scheme:
706 * Each feature has its own applicability flag that is set
707 * from csp->action->flags.
708 * Replaced the "filtering" int flag , by a function pointer
709 * "content_filter" to the function that will do the content
710 * modification. If it is != NULL, the document will be buffered
711 * and processed through *content_filter, which must set
712 * csp->content_length and return a modified copy of the body
713 * or return NULL (on failiure).
714 * - Changed csp->is_text to the more generic bitmap csp->content_type
715 * which can currently take the valued CT_TEXT or CT_GIF
717 * - Removed all #ifdef PCRS
719 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
720 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
721 * the `block_acl' call. This prevents a compilation error when the user
722 * does not wish to use the "ACL" feature.
724 * Revision 1.22 2001/06/29 21:45:41 oes
725 * Indentation, CRLF->LF, Tab-> Space
727 * Revision 1.21 2001/06/29 13:29:36 oes
728 * - Cleaned up, improved comments
729 * - Unified all possible interceptors (CGI,
730 * block, trust, fast_redirect) in one
731 * place, with one (CGI) answer generation
732 * mechansim. Much clearer now.
733 * - Removed the GIF image generation, which
734 * is now done in filters.c:block_url()
735 * - Made error conditions like domain lookup
736 * failiure or (various) problems while talking
737 * to the server use cgi.c:error_response()
738 * instead of generating HTML/HTTP in chat() (yuck!)
739 * - Removed logentry from cancelled commit
741 * Revision 1.20 2001/06/09 10:55:28 jongfoster
742 * Changing BUFSIZ ==> BUFFER_SIZE
744 * Revision 1.19 2001/06/07 23:12:52 jongfoster
745 * Replacing function pointer in struct gateway with a directly
746 * called function forwarded_connect().
747 * Replacing struct gateway with struct forward_spec
749 * Revision 1.18 2001/06/03 19:12:16 oes
750 * introduced new cgi handling
752 * Revision 1.17 2001/06/01 20:07:23 jongfoster
753 * Now uses action +image-blocker{} rather than config->tinygif
755 * Revision 1.16 2001/06/01 18:49:17 jongfoster
756 * Replaced "list_share" with "list" - the tiny memory gain was not
757 * worth the extra complexity.
759 * Revision 1.15 2001/05/31 21:24:47 jongfoster
760 * Changed "permission" to "action" throughout.
761 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
762 * Moved vanilla wafer check into chat(), since we must now
763 * decide whether or not to add it based on the URL.
765 * Revision 1.14 2001/05/29 20:14:01 joergs
766 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
769 * Revision 1.13 2001/05/29 09:50:24 jongfoster
770 * Unified blocklist/imagelist/permissionslist.
771 * File format is still under discussion, but the internal changes
774 * Also modified interceptor behaviour:
775 * - We now intercept all URLs beginning with one of the following
776 * prefixes (and *only* these prefixes):
778 * * http://ijbswa.sf.net/config/
779 * * http://ijbswa.sourceforge.net/config/
780 * - New interceptors "home page" - go to http://i.j.b/ to see it.
781 * - Internal changes so that intercepted and fast redirect pages
782 * are not replaced with an image.
783 * - Interceptors now have the option to send a binary page direct
784 * to the client. (i.e. ijb-send-banner uses this)
785 * - Implemented show-url-info interceptor. (Which is why I needed
786 * the above interceptors changes - a typical URL is
787 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
788 * The previous mechanism would not have intercepted that, and
789 * if it had been intercepted then it then it would have replaced
792 * Revision 1.12 2001/05/27 22:17:04 oes
794 * - re_process_buffer no longer writes the modified buffer
795 * to the client, which was very ugly. It now returns the
796 * buffer, which it is then written by chat.
798 * - content_length now adjusts the Content-Length: header
799 * for modified documents rather than crunch()ing it.
800 * (Length info in csp->content_length, which is 0 for
801 * unmodified documents)
803 * - For this to work, sed() is called twice when filtering.
805 * Revision 1.11 2001/05/26 17:27:53 jongfoster
806 * Added support for CLF and fixed LOG_LEVEL_LOG.
807 * Also did CRLF->LF fix of my previous patch.
809 * Revision 1.10 2001/05/26 15:26:15 jongfoster
810 * ACL feature now provides more security by immediately dropping
811 * connections from untrusted hosts.
813 * Revision 1.9 2001/05/26 00:28:36 jongfoster
814 * Automatic reloading of config file.
815 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
816 * Most of the global variables have been moved to a new
817 * struct configuration_spec, accessed through csp->config->globalname
818 * Most of the globals remaining are used by the Win32 GUI.
820 * Revision 1.8 2001/05/25 22:43:18 jongfoster
821 * Fixing minor memory leak and buffer overflow.
823 * Revision 1.7 2001/05/25 22:34:30 jongfoster
826 * Revision 1.6 2001/05/23 00:13:58 joergs
827 * AmigaOS support fixed.
829 * Revision 1.5 2001/05/22 18:46:04 oes
831 * - Enabled filtering banners by size rather than URL
832 * by adding patterns that replace all standard banner
833 * sizes with the "Junkbuster" gif to the re_filterfile
835 * - Enabled filtering WebBugs by providing a pattern
836 * which kills all 1x1 images
838 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
839 * which is selected by the (nonstandard and therefore
840 * capital) letter 'U' in the option string.
841 * It causes the quantifiers to be ungreedy by default.
842 * Appending a ? turns back to greedy (!).
844 * - Added a new interceptor ijb-send-banner, which
845 * sends back the "Junkbuster" gif. Without imagelist or
846 * MSIE detection support, or if tinygif = 1, or the
847 * URL isn't recognized as an imageurl, a lame HTML
848 * explanation is sent instead.
850 * - Added new feature, which permits blocking remote
851 * script redirects and firing back a local redirect
853 * The feature is conditionally compiled, i.e. it
854 * can be disabled with --disable-fast-redirects,
855 * plus it must be activated by a "fast-redirects"
856 * line in the config file, has its own log level
857 * and of course wants to be displayed by show-proxy-args
858 * Note: Boy, all the #ifdefs in 1001 locations and
859 * all the fumbling with configure.in and acconfig.h
860 * were *way* more work than the feature itself :-(
862 * - Because a generic redirect template was needed for
863 * this, tinygif = 3 now uses the same.
865 * - Moved GIFs, and other static HTTP response templates
870 * - Removed some >400 CRs again (Jon, you really worked
873 * Revision 1.4 2001/05/21 19:34:01 jongfoster
874 * Made failure to bind() a fatal error.
876 * Revision 1.3 2001/05/20 01:21:20 jongfoster
877 * Version 2.9.4 checkin.
878 * - Merged popupfile and cookiefile, and added control over PCRS
879 * filtering, in new "permissionsfile".
880 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
881 * file error you now get a message box (in the Win32 GUI) rather
882 * than the program exiting with no explanation.
883 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
885 * - Removed tabs from "config"
886 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
887 * - Bumped up version number.
889 * Revision 1.2 2001/05/17 22:34:44 oes
890 * - Added hint on GIF char array generation to jcc.c
891 * - Cleaned CRLF's from the sources and related files
892 * - Repaired logging for REF and FRC
894 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
895 * Initial import of version 2.9.3 source tree
898 *********************************************************************/
904 #include <sys/types.h>
913 # ifndef FEATURE_PTHREAD
917 # include <windows.h>
918 # include <process.h>
919 # endif /* ndef FEATURE_PTHREAD */
922 # ifndef _WIN_CONSOLE
924 # endif /* ndef _WIN_CONSOLE */
925 # include "w32svrapi.h"
927 #else /* ifndef _WIN32 */
929 # if !defined (__OS2__)
931 # include <sys/wait.h>
932 # endif /* ndef __OS2__ */
933 # include <sys/time.h>
934 # include <sys/stat.h>
935 # include <sys/ioctl.h>
938 #include <sys/termios.h>
949 # include <socket.h> /* BeOS has select() for sockets only. */
950 # include <OS.h> /* declarations for threads and stuff. */
953 # if defined(__EMX__) || defined(__OS2__)
954 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
959 #define bzero(B,N) memset(B,0x00,n)
974 #include "killpopup.h"
975 #include "miscutil.h"
977 #include "jbsockets.h"
982 #include "urlmatch.h"
984 const char jcc_h_rcs[] = JCC_H_VERSION;
985 const char project_h_rcs[] = PROJECT_H_VERSION;
988 struct client_state clients[1];
989 struct file_list files[1];
991 #ifdef FEATURE_STATISTICS
992 int urls_read = 0; /* total nr of urls read inc rejected */
993 int urls_rejected = 0; /* total nr of urls rejected */
994 #endif /* def FEATURE_STATISTICS */
996 #ifdef FEATURE_GRACEFUL_TERMINATION
1000 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1001 static void sig_handler(int the_signal);
1003 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1004 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1005 static jb_err get_server_headers(struct client_state *csp);
1006 static const char *crunch_reason(const struct http_response *rsp);
1007 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
1009 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1011 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1012 static jb_err change_request_destination(struct client_state *csp);
1013 static void chat(struct client_state *csp);
1014 static void serve(struct client_state *csp);
1016 static void usage(const char *myname);
1018 static void initialize_mutexes(void);
1019 static jb_socket bind_port_helper(struct configuration_spec *config);
1020 static void listen_loop(void);
1023 void serve(struct client_state *csp);
1024 #else /* ifndef AMIGA */
1025 static void serve(struct client_state *csp);
1026 #endif /* def AMIGA */
1029 static int32 server_thread(void *data);
1030 #endif /* def __BEOS__ */
1033 #define sleep(N) Sleep(((N) * 1000))
1037 #define sleep(N) DosSleep(((N) * 100))
1040 #ifdef FEATURE_PTHREAD
1041 pthread_mutex_t log_mutex;
1042 pthread_mutex_t log_init_mutex;
1044 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1045 pthread_mutex_t resolver_mutex;
1046 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1048 #ifndef HAVE_GMTIME_R
1049 pthread_mutex_t gmtime_mutex;
1050 #endif /* ndef HAVE_GMTIME_R */
1052 #ifndef HAVE_LOCALTIME_R
1053 pthread_mutex_t localtime_mutex;
1054 #endif /* ndef HAVE_GMTIME_R */
1057 pthread_mutex_t rand_mutex;
1058 #endif /* ndef HAVE_RANDOM */
1060 #endif /* FEATURE_PTHREAD */
1062 #if defined(unix) || defined(__EMX__)
1063 const char *basedir = NULL;
1064 const char *pidfile = NULL;
1065 int received_hup_signal = 0;
1066 #endif /* defined unix */
1068 /* HTTP snipplets. */
1069 static const char CSUCCEED[] =
1070 "HTTP/1.0 200 Connection established\n"
1071 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1073 static const char CHEADER[] =
1074 "HTTP/1.0 400 Invalid header received from browser\r\n"
1075 "Proxy-Agent: Privoxy " VERSION "\r\n"
1076 "Content-Type: text/plain\r\n"
1077 "Connection: close\r\n\r\n"
1078 "Invalid header received from browser.\r\n";
1080 static const char CFORBIDDEN[] =
1081 "HTTP/1.0 403 Connection not allowable\r\n"
1082 "Proxy-Agent: Privoxy " VERSION "\r\n"
1083 "X-Hint: If you read this message interactively, then you know why this happens ,-)\r\n"
1084 "Connection: close\r\n\r\n";
1086 static const char FTP_RESPONSE[] =
1087 "HTTP/1.0 400 Invalid request received from browser\r\n"
1088 "Content-Type: text/plain\r\n"
1089 "Connection: close\r\n\r\n"
1090 "Invalid request. Privoxy doesn't support FTP.\r\n";
1092 static const char GOPHER_RESPONSE[] =
1093 "HTTP/1.0 400 Invalid request received from browser\r\n"
1094 "Content-Type: text/plain\r\n"
1095 "Connection: close\r\n\r\n"
1096 "Invalid request. Privoxy doesn't support gopher.\r\n";
1098 /* XXX: should be a template */
1099 static const char MISSING_DESTINATION_RESPONSE[] =
1100 "HTTP/1.0 400 Bad request received from browser\r\n"
1101 "Proxy-Agent: Privoxy " VERSION "\r\n"
1102 "Content-Type: text/plain\r\n"
1103 "Connection: close\r\n\r\n"
1104 "Bad request. Privoxy was unable to extract the destination.\r\n";
1106 /* XXX: should be a template */
1107 static const char NO_SERVER_DATA_RESPONSE[] =
1108 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1109 "Proxy-Agent: Privoxy " VERSION "\r\n"
1110 "Content-Type: text/plain\r\n"
1111 "Connection: close\r\n\r\n"
1112 "Empty server or forwarder response.\r\n"
1113 "The connection was closed without sending any data.\r\n";
1116 /* XXX: should be a template */
1117 static const char NULL_BYTE_RESPONSE[] =
1118 "HTTP/1.0 400 Bad request received from browser\r\n"
1119 "Proxy-Agent: Privoxy " VERSION "\r\n"
1120 "Content-Type: text/plain\r\n"
1121 "Connection: close\r\n\r\n"
1122 "Bad request. Null byte(s) before end of request.\r\n";
1125 /* XXX: should be a template */
1126 static const char MESSED_UP_REQUEST_RESPONSE[] =
1127 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1128 "Proxy-Agent: Privoxy " VERSION "\r\n"
1129 "Content-Type: text/plain\r\n"
1130 "Connection: close\r\n\r\n"
1131 "Bad request. Messed up with header filters.\r\n";
1133 /* A function to crunch a response */
1134 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1136 /* Crunch function flags */
1137 #define CF_NO_FLAGS 0
1138 /* Cruncher applies to forced requests as well */
1139 #define CF_IGNORE_FORCE 1
1140 /* Crunched requests are counted for the block statistics */
1141 #define CF_COUNT_AS_REJECT 2
1143 /* A crunch function and its flags */
1146 const crunch_func_ptr cruncher;
1150 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1152 /* Complete list of cruncher functions */
1153 static const struct cruncher crunchers_all[] = {
1154 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1155 { block_url, CF_COUNT_AS_REJECT },
1156 #ifdef FEATURE_TRUST
1157 { trust_url, CF_COUNT_AS_REJECT },
1158 #endif /* def FEATURE_TRUST */
1159 { redirect_url, CF_NO_FLAGS },
1160 { dispatch_cgi, CF_IGNORE_FORCE},
1164 /* Light version, used after tags are applied */
1165 static const struct cruncher crunchers_light[] = {
1166 { block_url, CF_COUNT_AS_REJECT },
1167 { redirect_url, CF_NO_FLAGS },
1172 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1173 /*********************************************************************
1175 * Function : sig_handler
1177 * Description : Signal handler for different signals.
1178 * Exit gracefully on TERM and INT
1179 * or set a flag that will cause the errlog
1180 * to be reopened by the main thread on HUP.
1183 * 1 : the_signal = the signal cause this function to call
1187 *********************************************************************/
1188 static void sig_handler(int the_signal)
1194 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1205 received_hup_signal = 1;
1210 * We shouldn't be here, unless we catch signals
1211 * in main() that we can't handle here!
1213 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1221 /*********************************************************************
1223 * Function : client_protocol_is_unsupported
1225 * Description : Checks if the client used a known unsupported
1226 * protocol and deals with it by sending an error
1230 * 1 : csp = Current client state (buffers, headers, etc...)
1231 * 2 : req = the first request line send by the client
1233 * Returns : TRUE if an error response has been generated, or
1234 * FALSE if the request doesn't look invalid.
1236 *********************************************************************/
1237 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1239 char buf[BUFFER_SIZE];
1242 * If it's a FTP or gopher request, we don't support it.
1244 * These checks are better than nothing, but they might
1245 * not work in all configurations and some clients might
1246 * have problems digesting the answer.
1248 * They should, however, never cause more problems than
1249 * Privoxy's old behaviour (returning the misleading HTML
1252 * "Could not resolve http://(ftp|gopher)://example.org").
1254 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1256 if (!strncmpic(req, "GET ftp://", 10))
1258 strlcpy(buf, FTP_RESPONSE, sizeof(buf));
1259 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1260 csp->ip_addr_str, req);
1264 strlcpy(buf, GOPHER_RESPONSE, sizeof(buf));
1265 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1266 csp->ip_addr_str, req);
1268 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1270 write_socket(csp->cfd, buf, strlen(buf));
1279 /*********************************************************************
1281 * Function : get_request_destination_elsewhere
1283 * Description : If the client's request was redirected into
1284 * Privoxy without the client's knowledge,
1285 * the request line lacks the destination host.
1287 * This function tries to get it elsewhere,
1288 * provided accept-intercepted-requests is enabled.
1290 * "Elsewhere" currently only means "Host: header",
1291 * but in the future we may ask the redirecting
1292 * packet filter to look the destination up.
1294 * If the destination stays unknown, an error
1295 * response is send to the client and headers
1296 * are freed so that chat() can return directly.
1299 * 1 : csp = Current client state (buffers, headers, etc...)
1300 * 2 : headers = a header list
1302 * Returns : JB_ERR_OK if the destination is now known, or
1303 * JB_ERR_PARSE if it isn't.
1305 *********************************************************************/
1306 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1310 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1312 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1313 " Privoxy isn't configured to accept intercepted requests.",
1314 csp->ip_addr_str, csp->http->cmd);
1315 /* XXX: Use correct size */
1316 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1317 csp->ip_addr_str, csp->http->cmd);
1319 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1320 destroy_list(headers);
1322 return JB_ERR_PARSE;
1324 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1326 /* Split the domain we just got for pattern matching */
1327 init_domain_components(csp->http);
1333 /* We can't work without destination. Go spread the news.*/
1335 req = list_to_text(headers);
1337 /* XXX: Use correct size */
1338 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1339 csp->ip_addr_str, csp->http->cmd);
1340 log_error(LOG_LEVEL_ERROR,
1341 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1342 csp->ip_addr_str, csp->http->cmd, req);
1345 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1346 destroy_list(headers);
1348 return JB_ERR_PARSE;
1351 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1352 * to get the destination IP address, use it as host directly
1353 * or do a reverse DNS lookup first.
1358 /*********************************************************************
1360 * Function : get_server_headers
1362 * Description : Parses server headers in iob and fills them
1363 * into csp->headers so that they can later be
1367 * 1 : csp = Current client state (buffers, headers, etc...)
1369 * Returns : JB_ERR_OK if everything went fine, or
1370 * JB_ERR_PARSE if the headers were incomplete.
1372 *********************************************************************/
1373 static jb_err get_server_headers(struct client_state *csp)
1375 int continue_hack_in_da_house = 0;
1378 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1383 * continue hack in da house. Ignore the ending of
1384 * this head and continue enlisting header lines.
1385 * The reason is described below.
1387 enlist(csp->headers, "");
1388 continue_hack_in_da_house = 0;
1391 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1394 * It's a bodyless continue response, don't
1395 * stop header parsing after reaching it's end.
1397 * As a result Privoxy will concatenate the
1398 * next response's head and parse and deliver
1399 * the headers as if they belonged to one request.
1401 * The client will separate them because of the
1402 * empty line between them.
1404 * XXX: What we're doing here is clearly against
1405 * the intended purpose of the continue header,
1406 * and under some conditions (HTTP/1.0 client request)
1407 * it's a standard violation.
1409 * Anyway, "sort of against the spec" is preferable
1410 * to "always getting confused by Continue responses"
1411 * (Privoxy's behaviour before this hack was added)
1413 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1414 continue_hack_in_da_house = 1;
1416 else if (*header == '\0')
1419 * If the header is empty, but the Continue hack
1420 * isn't active, we can assume that we reached the
1421 * end of the buffer before we hit the end of the
1424 * Inform the caller an let it decide how to handle it.
1426 return JB_ERR_PARSE;
1430 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1433 * XXX: Should we quit the request and return a
1434 * out of memory error page instead?
1436 log_error(LOG_LEVEL_ERROR,
1437 "Out of memory while enlisting server headers. %s lost.",
1447 /*********************************************************************
1449 * Function : crunch_reason
1451 * Description : Translates the crunch reason code into a string.
1454 * 1 : rsp = a http_response
1456 * Returns : A string with the crunch reason or an error description.
1458 *********************************************************************/
1459 static const char *crunch_reason(const struct http_response *rsp)
1461 char * reason = NULL;
1463 assert(rsp != NULL);
1466 return "Internal error while searching for crunch reason";
1469 switch (rsp->reason)
1471 case RSP_REASON_UNSUPPORTED:
1472 reason = "Unsupported HTTP feature";
1474 case RSP_REASON_BLOCKED:
1477 case RSP_REASON_UNTRUSTED:
1478 reason = "Untrusted";
1480 case RSP_REASON_REDIRECTED:
1481 reason = "Redirected";
1483 case RSP_REASON_CGI_CALL:
1484 reason = "CGI Call";
1486 case RSP_REASON_NO_SUCH_DOMAIN:
1487 reason = "DNS failure";
1489 case RSP_REASON_FORWARDING_FAILED:
1490 reason = "Forwarding failed";
1492 case RSP_REASON_CONNECT_FAILED:
1493 reason = "Connection failure";
1495 case RSP_REASON_OUT_OF_MEMORY:
1496 reason = "Out of memory (may mask other reasons)";
1499 reason = "No reason recorded";
1507 /*********************************************************************
1509 * Function : send_crunch_response
1511 * Description : Delivers already prepared response for
1512 * intercepted requests, logs the interception
1513 * and frees the response.
1516 * 1 : csp = Current client state (buffers, headers, etc...)
1517 * 1 : rsp = Fully prepared response. Will be freed on exit.
1519 * Returns : Nothing.
1521 *********************************************************************/
1522 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1524 const struct http_request *http = csp->http;
1525 char status_code[4];
1527 assert(rsp != NULL);
1528 assert(rsp->head != NULL);
1533 * Not supposed to happen. If it does
1534 * anyway, treat it as an unknown error.
1536 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1537 /* return code doesn't matter */
1542 /* If rsp is still NULL, we have serious internal problems. */
1543 log_error(LOG_LEVEL_FATAL,
1544 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1548 * Extract the status code from the actual head
1549 * that was send to the client. It is the only
1550 * way to get it right for all requests, including
1551 * the fixed ones for out-of-memory problems.
1553 * A head starts like this: 'HTTP/1.1 200...'
1557 status_code[0] = rsp->head[9];
1558 status_code[1] = rsp->head[10];
1559 status_code[2] = rsp->head[11];
1560 status_code[3] = '\0';
1562 /* Write the answer to the client */
1563 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1564 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1566 /* There is nothing we can do about it. */
1567 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1570 /* Log that the request was crunched and why. */
1571 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1572 http->hostport, http->path, crunch_reason(rsp));
1573 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1574 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1576 /* Clean up and return */
1577 if (cgi_error_memory() != rsp)
1579 free_http_response(rsp);
1586 /*********************************************************************
1588 * Function : request_contains_null_bytes
1590 * Description : Checks for NULL bytes in the request and sends
1591 * an error message to the client if any were found.
1593 * XXX: currently not used, see comment in chat().
1596 * 1 : csp = Current client state (buffers, headers, etc...)
1597 * 2 : buf = Data from the client's request to check.
1598 * 3 : len = The data length.
1600 * Returns : TRUE if the request contained one or more NULL bytes, or
1603 *********************************************************************/
1604 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1606 size_t c_len; /* Request lenght when treated as C string */
1608 c_len = strlen(buf);
1613 * Null byte(s) found. Log the request,
1614 * return an error response and hang up.
1616 size_t tmp_len = c_len;
1621 * Replace NULL byte(s) with '°' characters
1622 * so the request can be logged as string.
1623 * XXX: Is there a better replacement character?
1626 tmp_len += strlen(buf+tmp_len);
1627 } while (tmp_len < len);
1629 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1630 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1631 log_error(LOG_LEVEL_HEADER,
1632 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1634 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1636 /* XXX: Log correct size */
1637 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1647 /*********************************************************************
1649 * Function : crunch_response_triggered
1651 * Description : Checks if the request has to be crunched,
1652 * and delivers the crunch response if necessary.
1655 * 1 : csp = Current client state (buffers, headers, etc...)
1656 * 2 : crunchers = list of cruncher functions to run
1658 * Returns : TRUE if the request was answered with a crunch response
1661 *********************************************************************/
1662 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1664 struct http_response *rsp = NULL;
1665 const struct cruncher *c;
1668 * If CGI request crunching is disabled,
1669 * check the CGI dispatcher out of order to
1670 * prevent unintentional blocks or redirects.
1672 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1673 && (NULL != (rsp = dispatch_cgi(csp))))
1675 /* Deliver, log and free the interception response. */
1676 send_crunch_response(csp, rsp);
1680 for (c = crunchers; c->cruncher != NULL; c++)
1683 * Check the cruncher if either Privoxy is toggled
1684 * on and the request isn't forced, or if the cruncher
1685 * applies to forced requests as well.
1687 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1688 !(csp->flags & CSP_FLAG_FORCED)) ||
1689 (c->flags & CF_IGNORE_FORCE))
1691 rsp = c->cruncher(csp);
1694 /* Deliver, log and free the interception response. */
1695 send_crunch_response(csp, rsp);
1696 #ifdef FEATURE_STATISTICS
1697 if (c->flags & CF_COUNT_AS_REJECT)
1699 csp->flags |= CSP_FLAG_REJECTED;
1701 #endif /* def FEATURE_STATISTICS */
1712 /*********************************************************************
1714 * Function : build_request_line
1716 * Description : Builds the HTTP request line.
1718 * If a HTTP forwarder is used it expects the whole URL,
1719 * web servers only get the path.
1722 * 1 : csp = Current client state (buffers, headers, etc...)
1723 * 2 : fwd = The forwarding spec used for the request
1724 * XXX: Should use http->fwd instead.
1725 * 3 : request_line = The old request line which will be replaced.
1727 * Returns : Nothing. Terminates in case of memory problems.
1729 *********************************************************************/
1730 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1732 struct http_request *http = csp->http;
1734 assert(http->ssl == 0);
1737 * Downgrade http version from 1.1 to 1.0
1738 * if +downgrade action applies.
1740 if ( (csp->action->flags & ACTION_DOWNGRADE)
1741 && (!strcmpic(http->ver, "HTTP/1.1")))
1744 http->ver = strdup("HTTP/1.0");
1746 if (http->ver == NULL)
1748 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1753 * Rebuild the request line.
1755 freez(*request_line);
1756 *request_line = strdup(http->gpc);
1757 string_append(request_line, " ");
1759 if (fwd->forward_host)
1761 string_append(request_line, http->url);
1765 string_append(request_line, http->path);
1767 string_append(request_line, " ");
1768 string_append(request_line, http->ver);
1770 if (*request_line == NULL)
1772 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1774 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1778 /*********************************************************************
1780 * Function : change_request_destination
1782 * Description : Parse a (rewritten) request line and regenerate
1783 * the http request data.
1786 * 1 : csp = Current client state (buffers, headers, etc...)
1788 * Returns : Forwards the parse_http_request() return code.
1789 * Terminates in case of memory problems.
1791 *********************************************************************/
1792 static jb_err change_request_destination(struct client_state *csp)
1794 struct http_request *http = csp->http;
1797 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1798 free_http_request(http);
1799 err = parse_http_request(csp->headers->first->str, http, csp);
1800 if (JB_ERR_OK != err)
1802 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1803 jb_err_to_string(err));
1805 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1806 if (http->ocmd == NULL)
1808 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1815 /*********************************************************************
1819 * Description : Once a connection to the client has been accepted,
1820 * this function is called (via serve()) to handle the
1821 * main business of the communication. When this
1822 * function returns, the caller must close the client
1825 * FIXME: chat is nearly thousand lines long.
1829 * 1 : csp = Current client state (buffers, headers, etc...)
1831 * Returns : Nothing.
1833 *********************************************************************/
1834 static void chat(struct client_state *csp)
1836 char buf[BUFFER_SIZE];
1844 int ms_iis5_hack = 0;
1845 size_t byte_count = 0;
1846 int forwarded_connect_retries = 0;
1847 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1848 const struct forward_spec * fwd;
1849 struct http_request *http;
1850 int len; /* for buffer sizes (and negative error codes) */
1852 #ifdef FEATURE_KILL_POPUPS
1853 int block_popups_now = 0; /* bool, 1==currently blocking popups */
1854 #endif /* def FEATURE_KILL_POPUPS */
1856 /* Function that does the content filtering for the current request */
1857 filter_function_ptr content_filter = NULL;
1859 /* Skeleton for HTTP response, if we should intercept the request */
1860 struct http_response *rsp;
1862 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1863 struct list header_list;
1864 struct list *headers = &header_list;
1868 memset(buf, 0, sizeof(buf));
1871 * Read the client's request. Note that since we're not using select() we
1872 * could get blocked here if a client connected, then didn't say anything!
1877 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1879 if (len <= 0) break; /* error! */
1882 * If there is no memory left for buffering the
1883 * request, there is nothing we can do but hang up
1885 if (add_to_iob(csp, buf, len))
1890 req = get_header(csp);
1892 } while ((NULL != req) && ('\0' == *req));
1894 if ((NULL != req) && ('\0' != *req))
1896 /* Request received. Validate and parse it. */
1900 * XXX: Temporary disabled to prevent problems
1901 * with POST requests whose bodies are allowed to
1902 * contain NULL bytes. BR#1730105.
1904 * The main purpose of this check is to properly
1905 * log stuff like BitTorrent traffic and other junk
1906 * that hits public proxies. It's not required for
1907 * Privoxy to functions as those requests are discarded
1910 * It probably should be rewritten to only check
1911 * the head of the request. Another option would
1912 * be to let all POST requests pass, although that
1913 * may not be good enough.
1915 if (request_contains_null_bytes(csp, buf, len))
1917 /* NULL bytes found and dealt with, just hang up. */
1922 /* Does the request line look invalid? */
1923 if (client_protocol_is_unsupported(csp, req))
1926 * Yes. The request has already been
1927 * answered with a error response, the buffers
1928 * were freed and we're done with chatting.
1933 #ifdef FEATURE_FORCE_LOAD
1935 * If this request contains the FORCE_PREFIX and blocks
1936 * aren't enforced, get rid of it and set the force flag.
1938 if (strstr(req, FORCE_PREFIX))
1940 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1942 log_error(LOG_LEVEL_FORCE,
1943 "Ignored force prefix in request: \"%s\".", req);
1947 strclean(req, FORCE_PREFIX);
1948 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
1949 csp->flags |= CSP_FLAG_FORCED;
1953 #endif /* def FEATURE_FORCE_LOAD */
1954 err = parse_http_request(req, http, csp);
1955 if (JB_ERR_OK != err)
1957 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
1963 if (http->cmd == NULL)
1965 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1966 /* XXX: Use correct size */
1967 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1968 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
1970 free_http_request(http);
1974 /* grab the rest of the client's headers */
1978 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
1980 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1983 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
1988 * If there is no memory left for buffering the
1989 * request, there is nothing we can do but hang up
1991 if (add_to_iob(csp, buf, len))
1998 if (p == NULL) break;
2005 if (http->host == NULL)
2008 * If we still don't know the request destination,
2009 * the request is invalid or the client uses
2010 * Privoxy without its knowledge.
2012 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2015 * Our attempts to get the request destination
2016 * elsewhere failed or Privoxy is configured
2017 * to only accept proxy requests.
2019 * An error response has already been send
2020 * and we're done here.
2027 * Determine the actions for this URL
2029 #ifdef FEATURE_TOGGLE
2030 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2032 /* Most compatible set of actions (i.e. none) */
2033 init_current_action(csp->action);
2036 #endif /* ndef FEATURE_TOGGLE */
2038 url_actions(http, csp);
2042 * Save a copy of the original request for logging
2044 http->ocmd = strdup(http->cmd);
2046 if (http->ocmd == NULL)
2048 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2051 enlist(csp->headers, http->cmd);
2053 /* Append the previously read headers */
2054 list_append_list_unique(csp->headers, headers);
2055 destroy_list(headers);
2057 err = sed(client_patterns, add_client_headers, csp);
2058 if (JB_ERR_OK != err)
2060 assert(err == JB_ERR_PARSE);
2061 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2063 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2065 /* Check request line for rewrites. */
2066 if ((NULL == csp->headers->first->str)
2067 || (strcmp(http->cmd, csp->headers->first->str) &&
2068 (JB_ERR_OK != change_request_destination(csp))))
2071 * A header filter broke the request line - bail out.
2073 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2074 /* XXX: Use correct size */
2075 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2076 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2078 free_http_request(http);
2082 /* decide how to route the HTTP request */
2083 if (NULL == (fwd = forward_url(http, csp)))
2085 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2086 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2090 * build the http request to send to the server
2091 * we have to do one of the following:
2093 * create = use the original HTTP request to create a new
2094 * HTTP request that has either the path component
2095 * without the http://domainspec (w/path) or the
2096 * full orininal URL (w/url)
2097 * Note that the path and/or the HTTP version may
2098 * have been altered by now.
2100 * connect = Open a socket to the host:port of the server
2101 * and short-circuit server and client socket.
2103 * pass = Pass the request unchanged if forwarding a CONNECT
2104 * request to a parent proxy. Note that we'll be sending
2105 * the CFAIL message ourselves if connecting to the parent
2106 * fails, but we won't send a CSUCCEED message if it works,
2107 * since that would result in a double message (ours and the
2108 * parent's). After sending the request to the parent, we simply
2111 * here's the matrix:
2114 * +--------+--------+
2116 * 0 | create | connect|
2118 * Forwarding +--------+--------+
2120 * 1 | create | pass |
2122 * +--------+--------+
2127 * Check if a CONNECT request is allowable:
2128 * In the absence of a +limit-connect action, allow only port 443.
2129 * If there is an action, allow whatever matches the specificaton.
2133 if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443)
2134 || (csp->action->flags & ACTION_LIMIT_CONNECT
2135 && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) )
2137 if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS)
2140 * The response may confuse some clients,
2141 * but makes unblocking easier.
2143 * XXX: It seems to work with all major browsers,
2144 * so we should consider returning a body by default someday ...
2146 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2147 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2148 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2150 csp->action->flags |= ACTION_BLOCK;
2155 write_socket(csp->cfd, CFORBIDDEN, strlen(CFORBIDDEN));
2156 log_error(LOG_LEVEL_INFO, "Request from %s denied. "
2157 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2158 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2160 assert(NULL != csp->http->ocmd);
2161 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 403 0", csp->ip_addr_str, csp->http->ocmd);
2163 list_remove_all(csp->headers);
2165 * XXX: For consistency we might want to log a crunch message here.
2174 freez(csp->headers->first->str);
2175 build_request_line(csp, fwd, &csp->headers->first->str);
2178 hdr = list_to_text(csp->headers);
2181 /* FIXME Should handle error properly */
2182 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2186 * We have a request. Check if one of the crunchers wants it.
2188 if (crunch_response_triggered(csp, crunchers_all))
2191 * Yes. The client got the crunch response
2192 * and we are done here after cleaning up.
2195 list_remove_all(csp->headers);
2201 * The headers can't be removed earlier because
2202 * they were still needed for the referrer check
2203 * in case of CGI crunches.
2205 * XXX: Would it be worth to move the referrer check
2206 * into client_referrer() and set a flag if it's trusted?
2208 list_remove_all(csp->headers);
2210 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2212 if (fwd->forward_host)
2214 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2215 fwd->forward_host, fwd->forward_port, http->hostport);
2219 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2222 /* here we connect to the server, gateway, or the forwarder */
2224 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2225 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2227 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2228 forwarded_connect_retries, http->hostport);
2231 if (csp->sfd == JB_INVALID_SOCKET)
2233 if (fwd->type != SOCKS_NONE)
2236 rsp = error_response(csp, "forwarding-failed", errno);
2238 else if (errno == EINVAL)
2240 rsp = error_response(csp, "no-such-domain", errno);
2244 rsp = error_response(csp, "connect-failed", errno);
2245 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2250 /* Write the answer to the client */
2253 send_crunch_response(csp, rsp);
2260 if (fwd->forward_host || (http->ssl == 0))
2262 /* write the client's (modified) header to the server
2263 * (along with anything else that may be in the buffer)
2266 if (write_socket(csp->sfd, hdr, strlen(hdr))
2267 || (flush_socket(csp->sfd, csp) < 0))
2269 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2272 rsp = error_response(csp, "connect-failed", errno);
2276 send_crunch_response(csp, rsp);
2286 * We're running an SSL tunnel and we're not forwarding,
2287 * so just send the "connect succeeded" message to the
2288 * client, flush the rest, and get out of the way.
2290 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2298 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2300 /* we're finished with the client's header */
2303 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2305 /* pass data between the client and server
2306 * until one or the other shuts down the connection.
2315 * FD_ZERO here seems to point to an errant macro which crashes.
2316 * So do this by hand for now...
2318 memset(&rfds,0x00,sizeof(fd_set));
2322 FD_SET(csp->cfd, &rfds);
2323 FD_SET(csp->sfd, &rfds);
2325 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2329 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2333 /* this is the body of the browser's request
2334 * just read it and write it.
2337 if (FD_ISSET(csp->cfd, &rfds))
2339 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2343 break; /* "game over, man" */
2346 if (write_socket(csp->sfd, buf, (size_t)len))
2348 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2355 * The server wants to talk. It could be the header or the body.
2356 * If `hdr' is null, then it's the header otherwise it's the body.
2357 * FIXME: Does `hdr' really mean `host'? No.
2361 if (FD_ISSET(csp->sfd, &rfds))
2364 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2368 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2370 if (http->ssl && (fwd->forward_host == NULL))
2373 * Just hang up. We already confirmed the client's CONNECT
2374 * request with status code 200 and unencrypted content is
2375 * no longer welcome.
2377 log_error(LOG_LEVEL_ERROR,
2378 "CONNECT already confirmed. Unable to tell the client about the problem.");
2381 else if (byte_count)
2384 * Just hang up. We already transmitted the original headers
2385 * and parts of the original content and therefore missed the
2386 * chance to send an error message (without risking data corruption).
2388 * XXX: we could retry with a fancy range request here.
2390 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2391 "Unable to tell the client about the problem.");
2395 rsp = error_response(csp, "connect-failed", errno);
2399 send_crunch_response(csp, rsp);
2405 /* Add a trailing zero. This lets filter_popups
2406 * use string operations.
2410 #ifdef FEATURE_KILL_POPUPS
2411 /* Filter the popups on this read. */
2412 if (block_popups_now)
2414 filter_popups(buf, csp);
2416 #endif /* def FEATURE_KILL_POPUPS */
2418 /* Normally, this would indicate that we've read
2419 * as much as the server has sent us and we can
2420 * close the client connection. However, Microsoft
2421 * in its wisdom has released IIS/5 with a bug that
2422 * prevents it from sending the trailing \r\n in
2423 * a 302 redirect header (and possibly other headers).
2424 * To work around this if we've haven't parsed
2425 * a full header we'll append a trailing \r\n
2426 * and see if this now generates a valid one.
2428 * This hack shouldn't have any impacts. If we've
2429 * already transmitted the header or if this is a
2430 * SSL connection, then we won't bother with this
2431 * hack. So we only work on partially received
2432 * headers. If we append a \r\n and this still
2433 * doesn't generate a valid header, then we won't
2434 * transmit anything to the client.
2439 if (server_body || http->ssl)
2442 * If we have been buffering up the document,
2443 * now is the time to apply content modification
2444 * and send the result to the client.
2448 p = execute_content_filter(csp, content_filter);
2450 * If the content filter fails, use the original
2451 * buffer and length.
2452 * (see p != NULL ? p : csp->iob->cur below)
2456 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2459 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2461 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2464 hdr = list_to_text(csp->headers);
2467 /* FIXME Should handle error properly */
2468 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2472 * Shouldn't happen because this was the second sed run
2473 * and tags are only created for the first one.
2475 assert(!crunch_response_triggered(csp, crunchers_all));
2477 if (write_socket(csp->cfd, hdr, strlen(hdr))
2478 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2480 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2490 break; /* "game over, man" */
2494 * This is NOT the body, so
2495 * Let's pretend the server just sent us a blank line.
2497 snprintf(buf, sizeof(buf), "\r\n");
2498 len = (int)strlen(buf);
2501 * Now, let the normal header parsing algorithm below do its
2502 * job. If it fails, we'll exit instead of continuing.
2509 * If this is an SSL connection or we're in the body
2510 * of the server document, just write it to the client,
2511 * unless we need to buffer the body for later content-filtering
2514 if (server_body || http->ssl)
2519 * If there is no memory left for buffering the content, or the buffer limit
2520 * has been reached, switch to non-filtering mode, i.e. make & write the
2521 * header, flush the iob and buf, and get out of the way.
2523 if (add_to_iob(csp, buf, len))
2528 log_error(LOG_LEVEL_ERROR, "Flushing header and buffers. Stepping back from filtering.");
2530 hdr = list_to_text(csp->headers);
2534 * Memory is too tight to even generate the header.
2535 * Send our static "Out-of-memory" page.
2537 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2538 rsp = cgi_error_memory();
2539 send_crunch_response(csp, rsp);
2543 hdrlen = strlen(hdr);
2545 if (write_socket(csp->cfd, hdr, hdrlen)
2546 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
2547 || (write_socket(csp->cfd, buf, (size_t)len)))
2549 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2556 * Reset the byte_count to the amount of bytes
2557 * we just flushed. len will be added a few lines below,
2558 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2560 byte_count = (size_t)flushed;
2562 content_filter = NULL;
2568 if (write_socket(csp->cfd, buf, (size_t)len))
2570 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2574 byte_count += (size_t)len;
2579 /* we're still looking for the end of the
2580 * server's header ... (does that make header
2581 * parsing an "out of body experience" ?
2585 * buffer up the data we just read. If that fails,
2586 * there's little we can do but send our static
2587 * out-of-memory page.
2589 if (add_to_iob(csp, buf, len))
2591 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2592 rsp = cgi_error_memory();
2593 send_crunch_response(csp, rsp);
2598 /* Convert iob into something sed() can digest */
2599 if (JB_ERR_PARSE == get_server_headers(csp))
2603 /* Well, we tried our MS IIS/5
2604 * hack and it didn't work.
2605 * The header is incomplete
2606 * and there isn't anything
2607 * we can do about it.
2613 /* Since we have to wait for
2614 * more from the server before
2615 * we can parse the headers
2616 * we just continue here.
2622 /* Did we actually get anything? */
2623 if (NULL == csp->headers->first)
2625 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2626 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2627 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2628 free_http_request(http);
2632 /* we have now received the entire header.
2633 * filter it and send the result to the client
2635 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2637 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2639 hdr = list_to_text(csp->headers);
2642 /* FIXME Should handle error properly */
2643 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2646 if (crunch_response_triggered(csp, crunchers_light))
2649 * One of the tags created by a server-header
2650 * tagger triggered a crunch. We already
2651 * delivered the crunch response to the client
2652 * and are done here after cleaning up.
2657 /* Buffer and pcrs filter this if appropriate. */
2659 if (!http->ssl) /* We talk plaintext */
2662 #ifdef FEATURE_KILL_POPUPS
2663 /* Start blocking popups if appropriate. */
2664 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2665 (csp->action->flags & ACTION_NO_POPUPS) != 0) /* Policy allows */
2667 block_popups_now = 1;
2669 * Filter the part of the body that came in the same read
2670 * as the last headers:
2672 filter_popups(csp->iob->cur, csp);
2674 #endif /* def FEATURE_KILL_POPUPS */
2675 content_filter = get_filter_function(csp);
2678 * Only write if we're not buffering for content modification
2680 if (!content_filter)
2682 /* write the server's (modified) header to
2683 * the client (along with anything else that
2684 * may be in the buffer)
2687 if (write_socket(csp->cfd, hdr, strlen(hdr))
2688 || ((len = flush_socket(csp->cfd, csp)) < 0))
2690 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2692 /* the write failed, so don't bother
2693 * mentioning it to the client...
2694 * it probably can't hear us anyway.
2700 byte_count += (size_t)len;
2703 /* we're finished with the server's header */
2708 /* If this was a MS IIS/5 hack then it means
2709 * the server has already closed the
2710 * connection. Nothing more to read. Time
2721 return; /* huh? we should never get here */
2724 if (csp->content_length == 0)
2727 * If Privoxy didn't recalculate the
2728 * Content-Lenght, byte_count is still
2731 csp->content_length = byte_count;
2734 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2735 csp->ip_addr_str, http->ocmd, csp->content_length);
2739 /*********************************************************************
2743 * Description : This is little more than chat. We only "serve" to
2744 * to close any socket that chat may have opened.
2747 * 1 : csp = Current client state (buffers, headers, etc...)
2751 *********************************************************************/
2753 void serve(struct client_state *csp)
2754 #else /* ifndef AMIGA */
2755 static void serve(struct client_state *csp)
2756 #endif /* def AMIGA */
2759 close_socket(csp->cfd);
2761 if (csp->sfd != JB_INVALID_SOCKET)
2763 close_socket(csp->sfd);
2766 csp->flags &= ~CSP_FLAG_ACTIVE;
2772 /*********************************************************************
2774 * Function : server_thread
2776 * Description : We only exist to call `serve' in a threaded environment.
2779 * 1 : data = Current client state (buffers, headers, etc...)
2781 * Returns : Always 0.
2783 *********************************************************************/
2784 static int32 server_thread(void *data)
2786 serve((struct client_state *) data);
2794 /*********************************************************************
2798 * Description : Print usage info & exit.
2800 * Parameters : Pointer to argv[0] for identifying ourselves
2804 *********************************************************************/
2805 static void usage(const char *myname)
2807 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2811 #endif /* defined(unix) */
2814 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2815 #endif /* defined(unix) */
2816 "[--version] [configfile]\n"
2817 "Aborting\n", myname);
2822 #endif /* defined(unix) */
2825 /*********************************************************************
2827 * Function : initialize_mutexes
2829 * Description : Prepares mutexes if mutex support is available.
2833 * Returns : Void, exits in case of errors.
2835 *********************************************************************/
2836 static void initialize_mutexes(void)
2840 #ifdef FEATURE_PTHREAD
2842 * Prepare global mutex semaphores
2844 err = pthread_mutex_init(&log_mutex, 0);
2846 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2849 * XXX: The assumptions below are a bit naive
2850 * and can cause locks that aren't necessary.
2852 * For example older FreeBSD versions (< 6.x?)
2853 * have no gethostbyname_r, but gethostbyname is
2856 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2857 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2858 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2860 * XXX: should we use a single mutex for
2861 * localtime() and gmtime() as well?
2863 #ifndef HAVE_GMTIME_R
2864 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2865 #endif /* ndef HAVE_GMTIME_R */
2867 #ifndef HAVE_LOCALTIME_R
2868 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2869 #endif /* ndef HAVE_GMTIME_R */
2872 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2873 #endif /* ndef HAVE_RANDOM */
2874 #endif /* FEATURE_PTHREAD */
2877 * TODO: mutex support for mingw32 would be swell.
2882 printf("Fatal error. Mutex initialization failed: %s.\n",
2891 /*********************************************************************
2895 * Description : Load the config file and start the listen loop.
2896 * This function is a lot more *sane* with the `load_config'
2897 * and `listen_loop' functions; although it stills does
2898 * a *little* too much for my taste.
2901 * 1 : argc = Number of parameters (including $0).
2902 * 2 : argv = Array of (char *)'s to the parameters.
2904 * Returns : 1 if : can't open config file, unrecognized directive,
2905 * stats requested in multi-thread mode, can't open the
2906 * log file, can't open the jar file, listen port is invalid,
2907 * any load fails, and can't bind port.
2909 * Else main never returns, the process must be signaled
2910 * to terminate execution. Or, on Windows, use the
2911 * "File", "Exit" menu option.
2913 *********************************************************************/
2915 int real_main(int argc, const char *argv[])
2917 int main(int argc, const char *argv[])
2921 unsigned int random_seed;
2923 struct passwd *pw = NULL;
2924 struct group *grp = NULL;
2927 char *pre_chroot_nslookup_to_load_resolver = NULL;
2934 #if !defined(_WIN32)
2942 * Parse the command line arguments
2944 * XXX: simply printing usage information in case of
2945 * invalid arguments isn't particular user friendly.
2947 while (++argc_pos < argc)
2950 /* Check to see if the service must be installed or uninstalled */
2951 if (strncmp(argv[argc_pos], "--install", 9) == 0)
2953 const char *pName = argv[argc_pos] + 9;
2956 exit( (install_service(pName)) ? 0 : 1 );
2958 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
2960 const char *pName = argv[argc_pos] + 11;
2963 exit((uninstall_service(pName)) ? 0 : 1);
2965 else if (strcmp(argv[argc_pos], "--service" ) == 0)
2967 bRunAsService = TRUE;
2968 w32_set_service_cwd();
2969 atexit(w32_service_exit_notify);
2972 #endif /* defined(_WIN32) */
2975 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2977 if (strcmp(argv[argc_pos], "--help") == 0)
2982 else if(strcmp(argv[argc_pos], "--version") == 0)
2984 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
2990 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
2995 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
2997 if (++argc_pos == argc) usage(argv[0]);
2998 pidfile = strdup(argv[argc_pos]);
3001 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3003 if (++argc_pos == argc) usage(argv[argc_pos]);
3005 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3008 if (NULL == (grp = getgrnam(p)))
3010 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3014 if (NULL == (pw = getpwnam(argv[argc_pos])))
3016 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3019 if (p != NULL) *--p = '\0';
3022 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3024 if (++argc_pos == argc) usage(argv[0]);
3025 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3028 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3032 #endif /* defined(unix) */
3034 else if (argc_pos + 1 != argc)
3037 * This is neither the last command line
3038 * option, nor was it recognized before,
3039 * therefore it must be invalid.
3045 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3047 configfile = argv[argc_pos];
3050 } /* -END- while (more arguments) */
3053 if ( *configfile != '/' )
3055 char cwd[BUFFER_SIZE];
3057 size_t abs_file_size;
3059 /* make config-filename absolute here */
3060 if (NULL == getcwd(cwd, sizeof(cwd)))
3062 perror("failed to get current working directory");
3067 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3068 basedir = strdup(cwd);
3070 if (NULL == basedir ||
3071 NULL == (abs_file = malloc(abs_file_size)))
3073 perror("malloc failed");
3076 strlcpy(abs_file, basedir, abs_file_size);
3077 strlcat(abs_file, "/", abs_file_size );
3078 strlcat(abs_file, configfile, abs_file_size);
3079 configfile = abs_file;
3081 #endif /* defined unix */
3085 clients->next = NULL;
3089 #elif defined(_WIN32)
3093 /* Prepare mutexes if supported and necessary. */
3094 initialize_mutexes();
3096 random_seed = (unsigned int)time(NULL);
3098 srandom(random_seed);
3101 #endif /* ifdef HAVE_RANDOM */
3104 * Unix signal handling
3106 * Catch the abort, interrupt and terminate signals for a graceful exit
3107 * Catch the hangup signal so the errlog can be reopened.
3108 * Ignore the broken pipe signals (FIXME: Why?)
3110 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3113 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3114 const int ignored_signals[] = { SIGPIPE, 0 };
3116 for (idx = 0; catched_signals[idx] != 0; idx++)
3118 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3119 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3121 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3122 #endif /* ifdef sun */
3124 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3128 for (idx = 0; ignored_signals[idx] != 0; idx++)
3130 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3132 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3137 #else /* ifdef _WIN32 */
3138 # ifdef _WIN_CONSOLE
3140 * We *are* in a windows console app.
3141 * Print a verbose messages about FAQ's and such
3143 printf("%s", win32_blurb);
3144 # endif /* def _WIN_CONSOLE */
3145 #endif /* def _WIN32 */
3148 /* Initialize the CGI subsystem */
3149 cgi_init_error_messages();
3152 * If runnig on unix and without the --nodaemon
3153 * option, become a daemon. I.e. fork, detach
3154 * from tty and get process group leadership
3167 if ( pid < 0 ) /* error */
3172 else if ( pid != 0 ) /* parent */
3177 * must check for errors
3178 * child died due to missing files aso
3181 wpid = waitpid( pid, &status, WNOHANG );
3190 /* Should be more portable, but not as well tested */
3195 #else /* ndef __FreeBSD__ */
3197 #endif /* ndef __FreeBSD__ */
3198 fd = open("/dev/tty", O_RDONLY);
3201 /* no error check here */
3202 ioctl( fd, TIOCNOTTY,0 );
3206 /* FIXME: should close stderr (fd 2) here too, but the test
3208 * and load config file is done in listen_loop() and puts
3209 * some messages on stderr there.
3216 } /* -END- if (!no_daemon) */
3219 * As soon as we have written the PID file, we can switch
3220 * to the user and group ID indicated by the --user option
3226 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3228 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3234 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3236 /* Read the time zone file from /etc before doing chroot. */
3238 if (NULL != pre_chroot_nslookup_to_load_resolver
3239 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3241 /* Initialize resolver library. */
3242 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3244 if (chroot(pw->pw_dir) < 0)
3246 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3250 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3253 if (setuid(pw->pw_uid))
3255 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3259 char putenv_dummy[64];
3261 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3262 if (putenv(putenv_dummy) != 0)
3264 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3267 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3268 if (putenv(putenv_dummy) != 0)
3270 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3276 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3279 #endif /* defined unix */
3282 /* This will be FALSE unless the command line specified --service
3286 /* Yup, so now we must attempt to establish a connection
3287 * with the service dispatcher. This will only work if this
3288 * process was launched by the service control manager to
3289 * actually run as a service. If this isn't the case, i've
3290 * known it take around 30 seconds or so for the call to return.
3293 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3294 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3296 /* Service has run, and at this point is now being stopped, so just return */
3301 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3303 /* An error occurred. Usually it's because --service was wrongly specified
3304 * and we were unable to connect to the Service Control Dispatcher because
3305 * it wasn't expecting us and is therefore not listening.
3307 * For now, just continue below to call the listen_loop function.
3310 #endif /* def _WIN32 */
3320 /*********************************************************************
3322 * Function : bind_port_helper
3324 * Description : Bind the listen port. Handles logging, and aborts
3328 * 1 : config = Privoxy configuration. Specifies port
3331 * Returns : Port that was opened.
3333 *********************************************************************/
3334 static jb_socket bind_port_helper(struct configuration_spec * config)
3339 if ( (config->haddr != NULL)
3340 && (config->haddr[0] == '1')
3341 && (config->haddr[1] == '2')
3342 && (config->haddr[2] == '7')
3343 && (config->haddr[3] == '.') )
3345 log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only",
3348 else if (config->haddr == NULL)
3350 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3355 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3356 config->hport, config->haddr);
3359 result = bind_port(config->haddr, config->hport, &bfd);
3366 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3367 "There may be another Privoxy or some other "
3368 "proxy running on port %d",
3369 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3370 config->hport, config->hport);
3373 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3374 "The hostname is not resolvable",
3375 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3378 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3379 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3382 /* shouldn't get here */
3383 return JB_INVALID_SOCKET;
3386 config->need_bind = 0;
3393 /* Without this simple workaround we get this compiler warning from _beginthread
3394 * warning C4028: formal parameter 1 different from declaration
3396 void w32_service_listen_loop(void *p)
3400 #endif /* def _WIN32 */
3403 /*********************************************************************
3405 * Function : listen_loop
3407 * Description : bind the listen port and enter a "FOREVER" listening loop.
3413 *********************************************************************/
3414 static void listen_loop(void)
3416 struct client_state *csp = NULL;
3418 struct configuration_spec * config;
3420 config = load_config();
3422 bfd = bind_port_helper(config);
3424 #ifdef FEATURE_GRACEFUL_TERMINATION
3425 while (!g_terminate)
3430 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3431 while (waitpid(-1, NULL, WNOHANG) > 0)
3433 /* zombie children */
3435 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3438 * Free data that was used by died threads
3444 * Re-open the errlog after HUP signal
3446 if (received_hup_signal)
3448 init_error_log(Argv[0], config->logfile, config->debug);
3449 received_hup_signal = 0;
3454 #ifdef FEATURE_COOKIE_JAR
3456 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3457 * already open. I think unload_configfile() is not being run, which should do
3458 * this work. Until that can get resolved, we'll use this workaround.
3462 if (csp->config->jar)
3464 fclose(csp->config->jar);
3465 csp->config->jar = NULL;
3467 #endif /* FEATURE_COOKIE_JAR */
3468 #endif /* __OS2__ */
3470 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3472 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3476 csp->flags |= CSP_FLAG_ACTIVE;
3477 csp->sfd = JB_INVALID_SOCKET;
3479 csp->config = config = load_config();
3481 if ( config->need_bind )
3484 * Since we were listening to the "old port", we will not see
3485 * a "listen" param change until the next IJB request. So, at
3486 * least 1 more request must be made for us to find the new
3487 * setting. I am simply closing the old socket and binding the
3490 * Which-ever is correct, we will serve 1 more page via the
3491 * old settings. This should probably be a "show-proxy-args"
3492 * request. This should not be a so common of an operation
3493 * that this will hurt people's feelings.
3498 bfd = bind_port_helper(config);
3501 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3503 if (!accept_connection(csp, bfd))
3505 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3518 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3521 #ifdef FEATURE_TOGGLE
3522 if (global_toggle_state)
3524 csp->flags |= CSP_FLAG_TOGGLED_ON;
3526 #endif /* def FEATURE_TOGGLE */
3528 if (run_loader(csp))
3530 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3531 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3535 if (block_acl(NULL,csp))
3537 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3538 close_socket(csp->cfd);
3542 #endif /* def FEATURE_ACL */
3544 /* add it to the list of clients */
3545 csp->next = clients->next;
3546 clients->next = csp;
3548 if (config->multi_threaded)
3552 /* this is a switch () statment in the C preprocessor - ugh */
3553 #undef SELECTED_ONE_OPTION
3555 /* Use Pthreads in preference to native code */
3556 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3557 #define SELECTED_ONE_OPTION
3559 pthread_t the_thread;
3560 pthread_attr_t attrs;
3562 pthread_attr_init(&attrs);
3563 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3564 errno = pthread_create(&the_thread, &attrs,
3565 (void * (*)(void *))serve, csp);
3566 child_id = errno ? -1 : 0;
3567 pthread_attr_destroy(&attrs);
3571 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3572 #define SELECTED_ONE_OPTION
3573 child_id = _beginthread(
3574 (void (*)(void *))serve,
3579 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3580 #define SELECTED_ONE_OPTION
3581 child_id = _beginthread(
3582 (void(* _Optlink)(void*))serve,
3588 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3589 #define SELECTED_ONE_OPTION
3591 thread_id tid = spawn_thread
3592 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3594 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3596 child_id = (int) tid;
3605 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3606 #define SELECTED_ONE_OPTION
3607 csp->cfd = ReleaseSocket(csp->cfd, -1);
3610 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3611 NP_Output, Output(),
3612 NP_CloseOutput, FALSE,
3613 NP_Name, (ULONG)"privoxy child",
3617 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3618 NP_Output, Output(),
3619 NP_CloseOutput, FALSE,
3620 NP_Name, (ULONG)"privoxy child",
3621 NP_StackSize, 200*1024,
3627 ((struct Task *)child_id)->tc_UserData = csp;
3628 Signal((struct Task *)child_id, SIGF_SINGLE);
3633 #if !defined(SELECTED_ONE_OPTION)
3636 /* This block is only needed when using fork().
3637 * When using threads, the server thread was
3638 * created and run by the call to _beginthread().
3640 if (child_id == 0) /* child */
3643 #ifdef FEATURE_TOGGLE
3644 int inherited_toggle_state = global_toggle_state;
3645 #endif /* def FEATURE_TOGGLE */
3650 * If we've been toggled or we've blocked the request, tell Mom
3653 #ifdef FEATURE_TOGGLE
3654 if (inherited_toggle_state != global_toggle_state)
3656 rc |= RC_FLAG_TOGGLED;
3658 #endif /* def FEATURE_TOGGLE */
3660 #ifdef FEATURE_STATISTICS
3661 if (csp->flags & CSP_FLAG_REJECTED)
3663 rc |= RC_FLAG_BLOCKED;
3665 #endif /* ndef FEATURE_STATISTICS */
3669 else if (child_id > 0) /* parent */
3671 /* in a fork()'d environment, the parent's
3672 * copy of the client socket and the CSP
3676 #if !defined(_WIN32) && !defined(__CYGWIN__)
3678 wait( &child_status );
3681 * Evaluate child's return code: If the child has
3682 * - been toggled, toggle ourselves
3683 * - blocked its request, bump up the stats counter
3686 #ifdef FEATURE_TOGGLE
3687 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3689 global_toggle_state = !global_toggle_state;
3691 #endif /* def FEATURE_TOGGLE */
3693 #ifdef FEATURE_STATISTICS
3695 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3699 #endif /* def FEATURE_STATISTICS */
3701 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3702 close_socket(csp->cfd);
3703 csp->flags &= ~CSP_FLAG_ACTIVE;
3707 #undef SELECTED_ONE_OPTION
3708 /* end of cpp switch () */
3710 if (child_id < 0) /* failed */
3712 char buf[BUFFER_SIZE];
3714 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3716 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3718 write_socket(csp->cfd, buf, strlen(buf));
3719 close_socket(csp->cfd);
3720 csp->flags &= ~CSP_FLAG_ACTIVE;
3731 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3733 /* Clean up. Aim: free all memory (no leaks) */
3734 #ifdef FEATURE_GRACEFUL_TERMINATION
3736 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3738 unload_current_config_file();
3739 unload_current_actions_file();
3740 unload_current_re_filterfile();
3741 #ifdef FEATURE_TRUST
3742 unload_current_trust_file();
3745 if (config->multi_threaded)
3752 } while ((clients->next != NULL) && (--i > 0));
3756 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3767 #ifdef FEATURE_COOKIE_JAR
3768 if (NULL != config->jar)
3770 fclose(config->jar);
3774 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3775 /* Cleanup - remove taskbar icon etc. */
3780 #endif /* FEATURE_GRACEFUL_TERMINATION */