1 const char jcc_rcs[] = "$Id: jcc.c,v 1.147 2007/08/25 14:42:40 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 1.147 2007/08/25 14:42:40 fabiankeil
37 * Don't crash if a broken header filter wiped out the request line.
39 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
40 * Fix byte_count calculation in case of flushes
41 * and don't parse the server headers a second time.
43 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
44 * - If there's a connection problem after we already forwarded
45 * parts of the original content, just hang up. Fixes BR#1776724.
46 * - Fix warnings about unused code on mingw32.
47 * - In case of flushes, calculate the byte count
48 * less incorrectly (I think).
50 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
51 * Add some more prototypes for static functions.
53 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
54 * Comment out request_contains_null_bytes() until it's used again.
56 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
57 * #1763173 from Stefan Huehner: s@const static@static const@
58 * and declare some more functions static.
60 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
61 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
62 * and explain why they were rejected in the first place.
63 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
64 * CONNECT requests. The request line was missing.
65 * - Add two more XXX reminders as we don't have enough already.
67 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
68 * As Hal noticed, checking dispatch_cgi() as the last cruncher
69 * looks like a bug if CGI requests are blocked unintentionally,
70 * so don't do it unless the user enabled the new config option
71 * "allow-cgi-request-crunching".
73 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
74 * - Allow to rewrite the request destination behind the client's back.
75 * - Turn the weird-looking unconditional for loop that
76 * reads the client request into a conditional while loop.
77 * Move the stuff that only runs once out of the loop.
78 * - Move parts of chat(), server_content_type() and the
79 * necessary stuff to fix BR#1750917 into get_filter_function().
81 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
82 * Temporary workaround for BR#1730105.
84 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
85 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
86 * deadlocks and crashes on OpenBSD and possibly other OS with neither
87 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
88 * Thanks to Ralf Horstmann for report and solution.
90 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
91 * Add forward-override{} to change the forwarding settings through
92 * action sections. This is mainly interesting to forward different
93 * clients differently (for example based on User-Agent or request
96 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
97 * - Let usage() mention the --chroot parameter.
98 * - Use read_socket() consistently and always leave
99 * the last buffer byte alone, even in cases where
100 * null termination (currently) doesn't matter.
102 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
103 * - Fix config file loading on Unix if no config file is specified.
104 * Since r1.97 Privoxy would always interpret the last argument as
105 * config file, even if it's a valid command line option.
106 * - Abort in case of unrecognized command line options. Closes #1719696.
107 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
108 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
110 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
111 * - Don't rerun crunchers that only depend on the request URL.
112 * - Don't count redirects and CGI requests as "blocked requests".
114 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
115 * Support crunching based on tags created by server-header taggers.
117 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
118 * Make HTTP snippets static (again). Add a Content-Type for those
119 * with content so the browser doesn't guess it based on the URL.
121 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
122 * Move crunching and request line rebuilding out of chat().
124 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
125 * Introduce tags as alternative way to specify which
126 * actions apply to a request. At the moment tags can be
127 * created based on client and server headers.
129 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
130 * Don't CLF-log CONNECT requests twice.
132 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
133 * Log the source address for ACL-related connection drops.
135 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
136 * New config option: enforce-blocks.
138 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
139 * - Move null byte check into separate function.
140 * - Don't confuse the client with error pages
141 * if a CONNECT request was already confirmed.
143 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
144 * Speed up NULL byte escaping and only log the complete
145 * NULL byte requests with header debugging enabled.
147 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
148 * Answer requests that contain NULL bytes with
149 * a custom response instead of waiting for more
150 * data until the client eventually hangs up.
152 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
153 * - Move delivery and logging of crunched responses
154 * from chat() into send_crunch_response().
155 * - Display the reason for generating http_responses.
156 * - Log the content length for LOG_LEVEL_CLF correctly
157 * (still incorrect for some fixed responses).
158 * - Reword an incorrect comment about
159 * treat-forbidden-connects-like-blocks violating
161 * - Add some log messages.
163 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
164 * Move mutex initialization into separate
165 * function and exit in case of errors.
167 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
168 * - Start to reduce chat()'s line count and move
169 * parts of it into separate functions.
170 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
172 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
173 * - Add Proxy-Agent header to HTTP snippets that are
174 * supposed to reach HTTP clients only.
175 * - Made a few CONNECT log messages more descriptive.
176 * - Catch completely empty server responses (as seen
177 * with Tor's fake ".noconnect" top level domain).
178 * - Use shiny new "forwarding-failed" template for socks errors.
180 * Revision 1.118 2007/01/07 07:43:43 joergs
181 * AmigaOS4 support added.
183 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
184 * Added config option accept-intercepted-requests
185 * and disabled it by default.
187 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
188 * Reverted parts of my last commit
189 * to keep error handling working.
191 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
192 * Fixed gcc43 conversion warnings.
194 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
195 * Fix -pedantic ISO C warning about converting
196 * from function pointer to object pointer.
198 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
199 * Silence compiler warning I introduced with my last commit.
201 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
202 * Mutex protect rand() if POSIX threading
203 * is used, warn the user if that's not possible
204 * and stop using it on _WIN32 where it could
207 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
208 * Don't prevent core dumps by catching SIGABRT.
209 * It's rude and makes debugging unreasonable painful.
211 * Revision 1.110 2006/12/13 14:52:53 etresoft
212 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
214 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
215 * Privoxy is now able to run as intercepting
216 * proxy in combination with any packet filter
217 * that does the port redirection. The destination
218 * is extracted from the "Host:" header which
219 * should be available for nearly all requests.
221 * Moved HTTP snipplets into jcc.c.
222 * Added error message for gopher proxy requests.
224 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
225 * Only unlink the pidfile if it's actually used.
227 * Change order of interception checks to make
228 * it possible to block or redirect requests for
231 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
232 * Make pthread mutex locking more generic. Instead of
233 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
234 * and use mutex locking unless there is an _r function
235 * available. Better safe than sorry.
237 * Fixes "./configure --disable-pthread" and should result
238 * in less threading-related problems on pthread-using platforms,
239 * but it still doesn't fix BR#1122404.
241 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
242 * Move pthread.h inclusion from jcc.c to jcc.h.
243 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
245 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
246 * Don't exit after receiving the second SIGHUP on Solaris.
248 * Fixes BR 1052235, but the same problem may exist on other
249 * systems. Once 3.0.6 is out we should use sigset()
250 * where available and see if it breaks anything.
252 * Revision 1.104 2006/09/23 13:26:38 roro
253 * Replace TABs by spaces in source code.
255 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
256 * Fix +redirect{}. Didn't work with -fast-redirects.
258 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
259 * Respond with 400 and a short text message
260 * if the client tries to use Privoxy as FTP proxy.
262 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
263 * Make number of retries in case of forwarded-connect problems
264 * a config file option (forwarded-connect-retries) and use 0 as
267 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
268 * Set random(3) seed.
270 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
271 * Follow the OpenBSD port's lead and protect the resolve
272 * functions on OpenBSD as well.
274 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
275 * --user fix. Only use the user as group if no group is specified.
276 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
278 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
279 * Windows service (re-)integration
283 * --install[:service_name]
284 * --uninstall[:service_name]
287 * They work as follows:
288 * --install will create a service for you and then terminate.
289 * By default the service name will be "privoxy" (without the quotes).
290 * However you can run multiple services if you wish, just by adding
291 * a colon and then a name (no spaces).
293 * --uninstall follows the exact same rules a --install.
295 * --service is used when the program is executed by the service
296 * control manager, and in normal circumstances would never be
297 * used as a command line argument.
299 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
300 * Windows service integration
302 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
303 * Incorporate Fabian Keil's patch work:
304 http://www.fabiankeil.de/sourcecode/privoxy/
306 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
307 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
308 * with what was really the latest development (the v_3_0_branch branch)
310 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
311 * Thanks to Jindrich Makovicka for a race condition fix for the log
312 * file. The race condition remains for non-pthread implementations.
313 * Reference patch #1175720.
315 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
316 * Add the ability to check jpeg images for invalid
317 * lengths of comment blocks. Defensive strategy
318 * against the exploit:
319 * Microsoft Security Bulletin MS04-028
320 * Buffer Overrun in JPEG Processing (GDI+) Could
321 * Allow Code Execution (833987)
322 * Enabled with +inspect-jpegs in actions files.
324 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
325 * - Fixed usage info for non-unix platforms
326 * - Fixed small cmdline parsing bug
328 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
329 * Diagnostics: Now preserve the returncode of pthread_create
330 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
332 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
333 * No longer ignore SIGCHLD. Fixes bug #769381
335 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
336 * Close jarfile on graceful exit, remove stray line
338 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
339 * Cosmetics: Killed a warning, a typo and an allocation left at exit
341 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
342 * No longer rely on non-POSIX.1 extensions of getcwd().
345 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
346 * Replaced setenv() by posix-compliant putenv()
347 * Thanks to Neil McCalden (nmcc AT users.sf.net).
349 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
350 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
352 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
353 * Clean-up and extension of improvements for forked mode:
354 * - Child's return code now consists of flags RC_FLAG_*
355 * - Reporting toggle to parent now properly #ifdef'ed
356 * - Children now report blocking to parent. This enables
357 * statistics in forked mode
359 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
360 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
361 * Children now report having been toggled through _exit(17), parents
362 * watch for that code and toggle themselves if found.
364 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
365 * Wrapping all *_r functions (the non-_r versions of them) with
366 * mutex semaphores for OSX. Hopefully this will take care of all
367 * of those pesky crash reports.
369 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
370 * Fixed two mostly harmless mem leaks
372 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
373 * Fix: Head of global clients list now initialized to NULL
375 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
376 * Added basic support for OPTIONS and TRACE HTTP methods:
377 * - New interceptor direct_response() added in chat().
378 * - sed() moved to earlier in the process, so that the
379 * Host: header is evaluated before actions and forwarding
382 * Revision 1.92 2002/05/08 16:00:46 oes
383 * Chat's buffer handling:
384 * - Fixed bug with unchecked out-of-mem conditions
385 * while reading client request & server headers
386 * - No longer predict if the buffer limit will be exceeded
387 * in the next read -- check add_to_iob's new
388 * return code. If buffer couldn't be extended
389 * (policy or out-of-mem) while
390 * - reading from client: abort
391 * - reading server headers: send error page
392 * - buffering server body for filter: flush,
393 * and if that fails: send error page
395 * Revision 1.91 2002/04/08 20:35:58 swa
398 * Revision 1.90 2002/04/02 14:57:28 oes
399 * Made sending wafers independent of FEATURE_COOKIE_JAR
401 * Revision 1.89 2002/03/31 17:18:59 jongfoster
402 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
404 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
405 * More compiler warning message maintenance
407 * Revision 1.87 2002/03/26 22:29:54 swa
408 * we have a new homepage!
410 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
411 * Workaround for closing the jarfile before load_config() comes around again
413 * Revision 1.85 2002/03/24 15:23:33 jongfoster
416 * Revision 1.84 2002/03/24 13:25:43 swa
417 * name change related issues
419 * Revision 1.83 2002/03/16 23:54:06 jongfoster
420 * Adding graceful termination feature, to help look for memory leaks.
421 * If you enable this (which, by design, has to be done by hand
422 * editing config.h) and then go to http://i.j.b/die, then the program
423 * will exit cleanly after the *next* request. It should free all the
424 * memory that was used.
426 * Revision 1.82 2002/03/13 00:27:05 jongfoster
429 * Revision 1.81 2002/03/12 01:42:50 oes
430 * Introduced modular filters
432 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
433 * OS/2 port maintenance:
434 * - Fixed EMX build - it had decayed a little
435 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
436 * substituted a memset for now.
438 * Revision 1.79 2002/03/09 20:03:52 jongfoster
439 * - Making various functions return int rather than size_t.
440 * (Undoing a recent change). Since size_t is unsigned on
441 * Windows, functions like read_socket that return -1 on
442 * error cannot return a size_t.
444 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
445 * crashes, and also frequently caused JB to jump to 100%
446 * CPU and stay there. (Because it thought it had just
447 * read ((unsigned)-1) == 4Gb of data...)
449 * - The signature of write_socket has changed, it now simply
450 * returns success=0/failure=nonzero.
452 * - Trying to get rid of a few warnings --with-debug on
453 * Windows, I've introduced a new type "jb_socket". This is
454 * used for the socket file descriptors. On Windows, this
455 * is SOCKET (a typedef for unsigned). Everywhere else, it's
456 * an int. The error value can't be -1 any more, so it's
457 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
458 * Windows it maps to the #define INVALID_SOCKET.)
460 * - The signature of bind_port has changed.
462 * Revision 1.78 2002/03/08 21:35:04 oes
463 * Added optional group supplement to --user option. Will now use default group of user if no group given
465 * Revision 1.77 2002/03/07 03:52:06 oes
466 * - Fixed compiler warnings etc
467 * - Improved handling of failed DNS lookups
469 * Revision 1.76 2002/03/06 22:54:35 jongfoster
470 * Automated function-comment nitpicking.
472 * Revision 1.75 2002/03/06 10:02:19 oes
473 * Fixed stupid bug when --user was not given
475 * Revision 1.74 2002/03/06 00:49:31 jongfoster
476 * Fixing warning on Windows
477 * Making #ifdefs that refer to the same variable consistently
478 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
480 * Revision 1.73 2002/03/05 23:57:30 hal9
481 * Stray character 's' on line 1618 was breaking build.
483 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
484 * - Re-enable OS/2 building after new parms were added
485 * - Fix false out of memory report when resolving CGI templates when no IP
486 * address is available of failed attempt (a la no such domain)
488 * Revision 1.71 2002/03/05 18:13:56 oes
489 * Added --user option
491 * Revision 1.70 2002/03/05 04:52:42 oes
492 * Deleted non-errlog debugging code
494 * Revision 1.69 2002/03/04 23:50:00 jongfoster
495 * Splitting off bind_port() call into bind_port_helper(), with
498 * Revision 1.68 2002/03/04 20:17:32 oes
501 * Revision 1.67 2002/03/04 18:18:57 oes
502 * - Removed _DEBUG mode
503 * - Cleand up cmdline parsing
504 * - Introduced --no-daemon, --pidfile options
505 * - Cleaned up signal handling:
506 * - Terminate cleanly on INT, TERM and ABRT
507 * - Schedule logfile for re-opening on HUP
508 * - Ignore CHLD and PIPE
509 * - Leave the rest with their default handlers
510 * - Uniform handler registration
511 * - Added usage() function
512 * - Played styleguide police
514 * Revision 1.66 2002/03/03 15:06:55 oes
515 * Re-enabled automatic config reloading
517 * Revision 1.65 2002/03/03 14:49:11 oes
518 * Fixed CLF logging: Now uses client's original HTTP request
520 * Revision 1.64 2002/03/03 09:18:03 joergs
521 * Made jumbjuster work on AmigaOS again.
523 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
524 * Clean up a little CRLF unpleasantness that suddenly appeared
526 * Revision 1.62 2002/02/20 23:17:23 jongfoster
527 * Detecting some out-of memory conditions and exiting with a log message.
529 * Revision 1.61 2002/01/17 21:01:52 jongfoster
530 * Moving all our URL and URL pattern parsing code to urlmatch.c.
532 * Revision 1.60 2001/12/30 14:07:32 steudten
533 * - Add signal handling (unix)
534 * - Add SIGHUP handler (unix)
535 * - Add creation of pidfile (unix)
536 * - Add action 'top' in rc file (RH)
537 * - Add entry 'SIGNALS' to manpage
538 * - Add exit message to logfile (unix)
540 * Revision 1.59 2001/12/13 14:07:18 oes
541 * Fixed Bug: 503 error page now sent OK
543 * Revision 1.58 2001/11/30 23:37:24 jongfoster
544 * Renaming the Win32 config file to config.txt - this is almost the
545 * same as the corresponding UNIX name "config"
547 * Revision 1.57 2001/11/16 00:47:43 jongfoster
548 * Changing the tty-disconnection code to use setsid().
550 * Revision 1.56 2001/11/13 20:20:54 jongfoster
551 * Tabs->spaces, fixing a bug with missing {} around an if()
553 * Revision 1.55 2001/11/13 20:14:53 jongfoster
554 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
556 * Revision 1.54 2001/11/07 00:03:14 steudten
557 * Give reliable return value if an error
558 * occurs not just 0 with new daemon mode.
560 * Revision 1.53 2001/11/05 21:41:43 steudten
561 * Add changes to be a real daemon just for unix os.
562 * (change cwd to /, detach from controlling tty, set
563 * process group and session leader to the own process.
565 * Add some fatal-error log message for failed malloc().
566 * Add '-d' if compiled with 'configure --with-debug' to
567 * enable debug output.
569 * Revision 1.52 2001/10/26 20:11:20 jongfoster
570 * Fixing type mismatch
572 * Revision 1.51 2001/10/26 17:38:28 oes
575 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
576 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
577 * threads to call select() simultaneously. So, it's time to do a real, live,
578 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
579 * (native). Both versions will work, but using __OS2__ offers multi-threading.
581 * Revision 1.49 2001/10/23 21:41:35 jongfoster
582 * Added call to initialize the (statically-allocated of course)
583 * "out of memory" CGI response.
585 * Revision 1.48 2001/10/10 19:56:46 jongfoster
586 * Moving some code that wasn't cookie-related out of an #ifdef
589 * Revision 1.47 2001/10/10 16:44:36 oes
590 * Added CONNECT destination port limitation check
592 * Revision 1.46 2001/10/08 15:17:41 oes
593 * Re-enabled SSL forwarding
595 * Revision 1.45 2001/10/07 15:42:11 oes
596 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
598 * Moved downgrading of the HTTP version from parse_http_request to
599 * chat(), since we can't decide if it is necessary before we have
600 * determined the actions for the URL. The HTTP command is now
601 * *always* re-built so the repairs need no longer be special-cased.
603 * filter_popups now gets a csp pointer so it can raise the new
604 * CSP_FLAG_MODIFIED flag.
608 * Added configurable size limit for the IOB. If the IOB grows so
609 * large that the next read would exceed the limit, the header
610 * is generated, and the header & unfiltered buffer are flushed
611 * to the client. Chat then continues in non-buffering,
612 * non-filtering body mode.
614 * Revision 1.44 2001/10/02 18:13:57 oes
617 * Revision 1.43 2001/10/02 15:32:13 oes
618 * Moved generation of hdr
620 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
621 * Cleaning up 2 compiler warnings on OS/2.
623 * Revision 1.41 2001/09/16 17:05:14 jongfoster
624 * Removing unused #include showarg.h
626 * Revision 1.40 2001/09/16 15:41:45 jongfoster
627 * Fixing signed/unsigned comparison warning.
629 * Revision 1.39 2001/09/16 13:21:27 jongfoster
630 * Changes to use new list functions.
632 * Revision 1.38 2001/09/16 13:01:46 jongfoster
633 * Removing redundant function call that zeroed zalloc()'d memory.
635 * Revision 1.37 2001/09/10 11:12:24 oes
636 * Deleted unused variable
638 * Revision 1.36 2001/09/10 10:56:15 oes
639 * Silenced compiler warnings
641 * Revision 1.35 2001/07/31 14:44:22 oes
642 * Deleted unused size parameter from filter_popups()
644 * Revision 1.34 2001/07/30 22:08:36 jongfoster
645 * Tidying up #defines:
646 * - All feature #defines are now of the form FEATURE_xxx
647 * - Permanently turned off WIN_GUI_EDIT
648 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
650 * Revision 1.33 2001/07/29 19:32:00 jongfoster
651 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
653 * Revision 1.32 2001/07/29 18:47:05 jongfoster
654 * Adding missing #include "loadcfg.h"
656 * Revision 1.31 2001/07/29 12:17:48 oes
657 * Applied pthread fix by Paul Lieverse
659 * Revision 1.30 2001/07/25 22:57:13 jongfoster
660 * __BEOS__ no longer overrides FEATURE_PTHREAD.
661 * This is because FEATURE_PTHREAD will soon be widely used, so I
662 * want to keep it simple.
664 * Revision 1.29 2001/07/24 12:47:06 oes
665 * Applied BeOS support update by Eugenia
667 * Revision 1.28 2001/07/23 13:26:12 oes
668 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
670 * Revision 1.27 2001/07/19 19:09:47 haroon
671 * - Added code to take care of the situation where while processing the first
672 * server response (which includes the server header), after finding the end
673 * of the headers we were not looking past the end of the headers for
674 * content modification. I enabled it for filter_popups.
675 * Someone else should look to see if other similar operations should be
676 * done to the discarded portion of the buffer.
678 * Note 2001/07/20: No, the other content modification mechanisms will process
679 * the whole iob later anyway. --oes
681 * Revision 1.26 2001/07/18 12:31:36 oes
684 * Revision 1.25 2001/07/15 19:43:49 jongfoster
685 * Supports POSIX threads.
686 * Also removed some unused #includes.
688 * Revision 1.24 2001/07/13 14:00:40 oes
689 * - Generic content modification scheme:
690 * Each feature has its own applicability flag that is set
691 * from csp->action->flags.
692 * Replaced the "filtering" int flag , by a function pointer
693 * "content_filter" to the function that will do the content
694 * modification. If it is != NULL, the document will be buffered
695 * and processed through *content_filter, which must set
696 * csp->content_length and return a modified copy of the body
697 * or return NULL (on failiure).
698 * - Changed csp->is_text to the more generic bitmap csp->content_type
699 * which can currently take the valued CT_TEXT or CT_GIF
701 * - Removed all #ifdef PCRS
703 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
704 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
705 * the `block_acl' call. This prevents a compilation error when the user
706 * does not wish to use the "ACL" feature.
708 * Revision 1.22 2001/06/29 21:45:41 oes
709 * Indentation, CRLF->LF, Tab-> Space
711 * Revision 1.21 2001/06/29 13:29:36 oes
712 * - Cleaned up, improved comments
713 * - Unified all possible interceptors (CGI,
714 * block, trust, fast_redirect) in one
715 * place, with one (CGI) answer generation
716 * mechansim. Much clearer now.
717 * - Removed the GIF image generation, which
718 * is now done in filters.c:block_url()
719 * - Made error conditions like domain lookup
720 * failiure or (various) problems while talking
721 * to the server use cgi.c:error_response()
722 * instead of generating HTML/HTTP in chat() (yuck!)
723 * - Removed logentry from cancelled commit
725 * Revision 1.20 2001/06/09 10:55:28 jongfoster
726 * Changing BUFSIZ ==> BUFFER_SIZE
728 * Revision 1.19 2001/06/07 23:12:52 jongfoster
729 * Replacing function pointer in struct gateway with a directly
730 * called function forwarded_connect().
731 * Replacing struct gateway with struct forward_spec
733 * Revision 1.18 2001/06/03 19:12:16 oes
734 * introduced new cgi handling
736 * Revision 1.17 2001/06/01 20:07:23 jongfoster
737 * Now uses action +image-blocker{} rather than config->tinygif
739 * Revision 1.16 2001/06/01 18:49:17 jongfoster
740 * Replaced "list_share" with "list" - the tiny memory gain was not
741 * worth the extra complexity.
743 * Revision 1.15 2001/05/31 21:24:47 jongfoster
744 * Changed "permission" to "action" throughout.
745 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
746 * Moved vanilla wafer check into chat(), since we must now
747 * decide whether or not to add it based on the URL.
749 * Revision 1.14 2001/05/29 20:14:01 joergs
750 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
753 * Revision 1.13 2001/05/29 09:50:24 jongfoster
754 * Unified blocklist/imagelist/permissionslist.
755 * File format is still under discussion, but the internal changes
758 * Also modified interceptor behaviour:
759 * - We now intercept all URLs beginning with one of the following
760 * prefixes (and *only* these prefixes):
762 * * http://ijbswa.sf.net/config/
763 * * http://ijbswa.sourceforge.net/config/
764 * - New interceptors "home page" - go to http://i.j.b/ to see it.
765 * - Internal changes so that intercepted and fast redirect pages
766 * are not replaced with an image.
767 * - Interceptors now have the option to send a binary page direct
768 * to the client. (i.e. ijb-send-banner uses this)
769 * - Implemented show-url-info interceptor. (Which is why I needed
770 * the above interceptors changes - a typical URL is
771 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
772 * The previous mechanism would not have intercepted that, and
773 * if it had been intercepted then it then it would have replaced
776 * Revision 1.12 2001/05/27 22:17:04 oes
778 * - re_process_buffer no longer writes the modified buffer
779 * to the client, which was very ugly. It now returns the
780 * buffer, which it is then written by chat.
782 * - content_length now adjusts the Content-Length: header
783 * for modified documents rather than crunch()ing it.
784 * (Length info in csp->content_length, which is 0 for
785 * unmodified documents)
787 * - For this to work, sed() is called twice when filtering.
789 * Revision 1.11 2001/05/26 17:27:53 jongfoster
790 * Added support for CLF and fixed LOG_LEVEL_LOG.
791 * Also did CRLF->LF fix of my previous patch.
793 * Revision 1.10 2001/05/26 15:26:15 jongfoster
794 * ACL feature now provides more security by immediately dropping
795 * connections from untrusted hosts.
797 * Revision 1.9 2001/05/26 00:28:36 jongfoster
798 * Automatic reloading of config file.
799 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
800 * Most of the global variables have been moved to a new
801 * struct configuration_spec, accessed through csp->config->globalname
802 * Most of the globals remaining are used by the Win32 GUI.
804 * Revision 1.8 2001/05/25 22:43:18 jongfoster
805 * Fixing minor memory leak and buffer overflow.
807 * Revision 1.7 2001/05/25 22:34:30 jongfoster
810 * Revision 1.6 2001/05/23 00:13:58 joergs
811 * AmigaOS support fixed.
813 * Revision 1.5 2001/05/22 18:46:04 oes
815 * - Enabled filtering banners by size rather than URL
816 * by adding patterns that replace all standard banner
817 * sizes with the "Junkbuster" gif to the re_filterfile
819 * - Enabled filtering WebBugs by providing a pattern
820 * which kills all 1x1 images
822 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
823 * which is selected by the (nonstandard and therefore
824 * capital) letter 'U' in the option string.
825 * It causes the quantifiers to be ungreedy by default.
826 * Appending a ? turns back to greedy (!).
828 * - Added a new interceptor ijb-send-banner, which
829 * sends back the "Junkbuster" gif. Without imagelist or
830 * MSIE detection support, or if tinygif = 1, or the
831 * URL isn't recognized as an imageurl, a lame HTML
832 * explanation is sent instead.
834 * - Added new feature, which permits blocking remote
835 * script redirects and firing back a local redirect
837 * The feature is conditionally compiled, i.e. it
838 * can be disabled with --disable-fast-redirects,
839 * plus it must be activated by a "fast-redirects"
840 * line in the config file, has its own log level
841 * and of course wants to be displayed by show-proxy-args
842 * Note: Boy, all the #ifdefs in 1001 locations and
843 * all the fumbling with configure.in and acconfig.h
844 * were *way* more work than the feature itself :-(
846 * - Because a generic redirect template was needed for
847 * this, tinygif = 3 now uses the same.
849 * - Moved GIFs, and other static HTTP response templates
854 * - Removed some >400 CRs again (Jon, you really worked
857 * Revision 1.4 2001/05/21 19:34:01 jongfoster
858 * Made failure to bind() a fatal error.
860 * Revision 1.3 2001/05/20 01:21:20 jongfoster
861 * Version 2.9.4 checkin.
862 * - Merged popupfile and cookiefile, and added control over PCRS
863 * filtering, in new "permissionsfile".
864 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
865 * file error you now get a message box (in the Win32 GUI) rather
866 * than the program exiting with no explanation.
867 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
869 * - Removed tabs from "config"
870 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
871 * - Bumped up version number.
873 * Revision 1.2 2001/05/17 22:34:44 oes
874 * - Added hint on GIF char array generation to jcc.c
875 * - Cleaned CRLF's from the sources and related files
876 * - Repaired logging for REF and FRC
878 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
879 * Initial import of version 2.9.3 source tree
882 *********************************************************************/
888 #include <sys/types.h>
897 # ifndef FEATURE_PTHREAD
901 # include <windows.h>
902 # include <process.h>
903 # endif /* ndef FEATURE_PTHREAD */
906 # ifndef _WIN_CONSOLE
908 # endif /* ndef _WIN_CONSOLE */
909 # include "w32svrapi.h"
911 #else /* ifndef _WIN32 */
913 # if !defined (__OS2__)
915 # include <sys/wait.h>
916 # endif /* ndef __OS2__ */
917 # include <sys/time.h>
918 # include <sys/stat.h>
919 # include <sys/ioctl.h>
922 #include <sys/termios.h>
933 # include <socket.h> /* BeOS has select() for sockets only. */
934 # include <OS.h> /* declarations for threads and stuff. */
937 # if defined(__EMX__) || defined(__OS2__)
938 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
943 #define bzero(B,N) memset(B,0x00,n)
958 #include "killpopup.h"
959 #include "miscutil.h"
961 #include "jbsockets.h"
966 #include "urlmatch.h"
968 const char jcc_h_rcs[] = JCC_H_VERSION;
969 const char project_h_rcs[] = PROJECT_H_VERSION;
972 struct client_state clients[1];
973 struct file_list files[1];
975 #ifdef FEATURE_STATISTICS
976 int urls_read = 0; /* total nr of urls read inc rejected */
977 int urls_rejected = 0; /* total nr of urls rejected */
978 #endif /* def FEATURE_STATISTICS */
980 #ifdef FEATURE_GRACEFUL_TERMINATION
984 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
985 static void sig_handler(int the_signal);
987 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
988 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
989 static jb_err get_server_headers(struct client_state *csp);
990 static const char *crunch_reason(const struct http_response *rsp);
991 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
993 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
995 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
996 static jb_err change_request_destination(struct client_state *csp);
997 static void chat(struct client_state *csp);
998 static void serve(struct client_state *csp);
1000 static void usage(const char *myname);
1002 static void initialize_mutexes(void);
1003 static jb_socket bind_port_helper(struct configuration_spec *config);
1004 static void listen_loop(void);
1007 void serve(struct client_state *csp);
1008 #else /* ifndef AMIGA */
1009 static void serve(struct client_state *csp);
1010 #endif /* def AMIGA */
1013 static int32 server_thread(void *data);
1014 #endif /* def __BEOS__ */
1017 #define sleep(N) Sleep(((N) * 1000))
1021 #define sleep(N) DosSleep(((N) * 100))
1024 #ifdef FEATURE_PTHREAD
1025 pthread_mutex_t log_mutex;
1026 pthread_mutex_t log_init_mutex;
1028 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1029 pthread_mutex_t resolver_mutex;
1030 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1032 #ifndef HAVE_GMTIME_R
1033 pthread_mutex_t gmtime_mutex;
1034 #endif /* ndef HAVE_GMTIME_R */
1036 #ifndef HAVE_LOCALTIME_R
1037 pthread_mutex_t localtime_mutex;
1038 #endif /* ndef HAVE_GMTIME_R */
1041 pthread_mutex_t rand_mutex;
1042 #endif /* ndef HAVE_RANDOM */
1044 #endif /* FEATURE_PTHREAD */
1046 #if defined(unix) || defined(__EMX__)
1047 const char *basedir = NULL;
1048 const char *pidfile = NULL;
1049 int received_hup_signal = 0;
1050 #endif /* defined unix */
1052 /* The vanilla wafer. */
1053 static const char VANILLA_WAFER[] =
1054 "NOTICE=TO_WHOM_IT_MAY_CONCERN_"
1055 "Do_not_send_me_any_copyrighted_information_other_than_the_"
1056 "document_that_I_am_requesting_or_any_of_its_necessary_components._"
1057 "In_particular_do_not_send_me_any_cookies_that_"
1058 "are_subject_to_a_claim_of_copyright_by_anybody._"
1059 "Take_notice_that_I_refuse_to_be_bound_by_any_license_condition_"
1060 "(copyright_or_otherwise)_applying_to_any_cookie._";
1062 /* HTTP snipplets. */
1063 static const char CSUCCEED[] =
1064 "HTTP/1.0 200 Connection established\n"
1065 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1067 static const char CHEADER[] =
1068 "HTTP/1.0 400 Invalid header received from browser\r\n"
1069 "Proxy-Agent: Privoxy " VERSION "\r\n"
1070 "Content-Type: text/plain\r\n"
1071 "Connection: close\r\n\r\n"
1072 "Invalid header received from browser.\r\n";
1074 static const char CFORBIDDEN[] =
1075 "HTTP/1.0 403 Connection not allowable\r\n"
1076 "Proxy-Agent: Privoxy " VERSION "\r\n"
1077 "X-Hint: If you read this message interactively, then you know why this happens ,-)\r\n"
1078 "Connection: close\r\n\r\n";
1080 static const char FTP_RESPONSE[] =
1081 "HTTP/1.0 400 Invalid request received from browser\r\n"
1082 "Content-Type: text/plain\r\n"
1083 "Connection: close\r\n\r\n"
1084 "Invalid request. Privoxy doesn't support FTP.\r\n";
1086 static const char GOPHER_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 gopher.\r\n";
1092 /* XXX: should be a template */
1093 static const char MISSING_DESTINATION_RESPONSE[] =
1094 "HTTP/1.0 400 Bad request received from browser\r\n"
1095 "Proxy-Agent: Privoxy " VERSION "\r\n"
1096 "Content-Type: text/plain\r\n"
1097 "Connection: close\r\n\r\n"
1098 "Bad request. Privoxy was unable to extract the destination.\r\n";
1100 /* XXX: should be a template */
1101 static const char NO_SERVER_DATA_RESPONSE[] =
1102 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1103 "Proxy-Agent: Privoxy " VERSION "\r\n"
1104 "Content-Type: text/plain\r\n"
1105 "Connection: close\r\n\r\n"
1106 "Empty server or forwarder response.\r\n"
1107 "The connection was closed without sending any data.\r\n";
1109 /* XXX: should be a template */
1110 static const char NULL_BYTE_RESPONSE[] =
1111 "HTTP/1.0 400 Bad request received from browser\r\n"
1112 "Proxy-Agent: Privoxy " VERSION "\r\n"
1113 "Content-Type: text/plain\r\n"
1114 "Connection: close\r\n\r\n"
1115 "Bad request. Null byte(s) before end of request.\r\n";
1117 /* XXX: should be a template */
1118 static const char MESSED_UP_REQUEST_RESPONSE[] =
1119 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1120 "Proxy-Agent: Privoxy " VERSION "\r\n"
1121 "Content-Type: text/plain\r\n"
1122 "Connection: close\r\n\r\n"
1123 "Bad request. Messed up with header filters.\r\n";
1125 /* A function to crunch a response */
1126 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1128 typedef char *(*filter_function_ptr)();
1130 /* Crunch function flags */
1131 #define CF_NO_FLAGS 0
1132 /* Cruncher applies to forced requests as well */
1133 #define CF_IGNORE_FORCE 1
1134 /* Crunched requests are counted for the block statistics */
1135 #define CF_COUNT_AS_REJECT 2
1137 /* A crunch function and its flags */
1140 const crunch_func_ptr cruncher;
1144 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1145 static filter_function_ptr get_filter_function(struct client_state *csp);
1147 /* Complete list of cruncher functions */
1148 static const struct cruncher crunchers_all[] = {
1149 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1150 { block_url, CF_COUNT_AS_REJECT },
1151 #ifdef FEATURE_TRUST
1152 { trust_url, CF_COUNT_AS_REJECT },
1153 #endif /* def FEATURE_TRUST */
1154 { redirect_url, CF_NO_FLAGS },
1155 { dispatch_cgi, CF_IGNORE_FORCE},
1159 /* Light version, used after tags are applied */
1160 static const struct cruncher crunchers_light[] = {
1161 { block_url, CF_COUNT_AS_REJECT },
1162 { redirect_url, CF_NO_FLAGS },
1167 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1168 /*********************************************************************
1170 * Function : sig_handler
1172 * Description : Signal handler for different signals.
1173 * Exit gracefully on TERM and INT
1174 * or set a flag that will cause the errlog
1175 * to be reopened by the main thread on HUP.
1178 * 1 : the_signal = the signal cause this function to call
1182 *********************************************************************/
1183 static void sig_handler(int the_signal)
1189 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1200 received_hup_signal = 1;
1205 * We shouldn't be here, unless we catch signals
1206 * in main() that we can't handle here!
1208 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1216 /*********************************************************************
1218 * Function : client_protocol_is_unsupported
1220 * Description : Checks if the client used a known unsupported
1221 * protocol and deals with it by sending an error
1225 * 1 : csp = Current client state (buffers, headers, etc...)
1226 * 2 : req = the first request line send by the client
1228 * Returns : TRUE if an error response has been generated, or
1229 * FALSE if the request doesn't look invalid.
1231 *********************************************************************/
1232 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1234 char buf[BUFFER_SIZE];
1237 * If it's a FTP or gopher request, we don't support it.
1239 * These checks are better than nothing, but they might
1240 * not work in all configurations and some clients might
1241 * have problems digesting the answer.
1243 * They should, however, never cause more problems than
1244 * Privoxy's old behaviour (returning the misleading HTML
1247 * "Could not resolve http://(ftp|gopher)://example.org").
1249 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1251 if (!strncmpic(req, "GET ftp://", 10))
1253 strlcpy(buf, FTP_RESPONSE, sizeof(buf));
1254 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1255 csp->ip_addr_str, req);
1259 strlcpy(buf, GOPHER_RESPONSE, sizeof(buf));
1260 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1261 csp->ip_addr_str, req);
1263 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1265 write_socket(csp->cfd, buf, strlen(buf));
1274 /*********************************************************************
1276 * Function : get_request_destination_elsewhere
1278 * Description : If the client's request was redirected into
1279 * Privoxy without the client's knowledge,
1280 * the request line lacks the destination host.
1282 * This function tries to get it elsewhere,
1283 * provided accept-intercepted-requests is enabled.
1285 * "Elsewhere" currently only means "Host: header",
1286 * but in the future we may ask the redirecting
1287 * packet filter to look the destination up.
1289 * If the destination stays unknown, an error
1290 * response is send to the client and headers
1291 * are freed so that chat() can return directly.
1294 * 1 : csp = Current client state (buffers, headers, etc...)
1295 * 2 : headers = a header list
1297 * Returns : JB_ERR_OK if the destination is now known, or
1298 * JB_ERR_PARSE if it isn't.
1300 *********************************************************************/
1301 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1305 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1307 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1308 " Privoxy isn't configured to accept intercepted requests.",
1309 csp->ip_addr_str, csp->http->cmd);
1310 /* XXX: Use correct size */
1311 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1312 csp->ip_addr_str, csp->http->cmd);
1314 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1315 destroy_list(headers);
1317 return JB_ERR_PARSE;
1319 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1321 /* Split the domain we just got for pattern matching */
1322 init_domain_components(csp->http);
1328 /* We can't work without destination. Go spread the news.*/
1330 req = list_to_text(headers);
1332 /* XXX: Use correct size */
1333 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1334 csp->ip_addr_str, csp->http->cmd);
1335 log_error(LOG_LEVEL_ERROR,
1336 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1337 csp->ip_addr_str, csp->http->cmd, req);
1340 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1341 destroy_list(headers);
1343 return JB_ERR_PARSE;
1346 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1347 * to get the destination IP address, use it as host directly
1348 * or do a reverse DNS lookup first.
1353 /*********************************************************************
1355 * Function : get_server_headers
1357 * Description : Parses server headers in iob and fills them
1358 * into csp->headers so that they can later be
1362 * 1 : csp = Current client state (buffers, headers, etc...)
1364 * Returns : JB_ERR_OK if everything went fine, or
1365 * JB_ERR_PARSE if the headers were incomplete.
1367 *********************************************************************/
1368 static jb_err get_server_headers(struct client_state *csp)
1370 int continue_hack_in_da_house = 0;
1373 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1378 * continue hack in da house. Ignore the ending of
1379 * this head and continue enlisting header lines.
1380 * The reason is described below.
1382 enlist(csp->headers, "");
1383 continue_hack_in_da_house = 0;
1386 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1389 * It's a bodyless continue response, don't
1390 * stop header parsing after reaching it's end.
1392 * As a result Privoxy will concatenate the
1393 * next response's head and parse and deliver
1394 * the headers as if they belonged to one request.
1396 * The client will separate them because of the
1397 * empty line between them.
1399 * XXX: What we're doing here is clearly against
1400 * the intended purpose of the continue header,
1401 * and under some conditions (HTTP/1.0 client request)
1402 * it's a standard violation.
1404 * Anyway, "sort of against the spec" is preferable
1405 * to "always getting confused by Continue responses"
1406 * (Privoxy's behaviour before this hack was added)
1408 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1409 continue_hack_in_da_house = 1;
1411 else if (*header == '\0')
1414 * If the header is empty, but the Continue hack
1415 * isn't active, we can assume that we reached the
1416 * end of the buffer before we hit the end of the
1419 * Inform the caller an let it decide how to handle it.
1421 return JB_ERR_PARSE;
1425 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1428 * XXX: Should we quit the request and return a
1429 * out of memory error page instead?
1431 log_error(LOG_LEVEL_ERROR,
1432 "Out of memory while enlisting server headers. %s lost.",
1442 /*********************************************************************
1444 * Function : crunch_reason
1446 * Description : Translates the crunch reason code into a string.
1449 * 1 : rsp = a http_response
1451 * Returns : A string with the crunch reason or an error description.
1453 *********************************************************************/
1454 static const char *crunch_reason(const struct http_response *rsp)
1456 char * reason = NULL;
1458 assert(rsp != NULL);
1461 return "Internal error while searching for crunch reason";
1464 switch (rsp->reason)
1466 case RSP_REASON_UNSUPPORTED:
1467 reason = "Unsupported HTTP feature";
1469 case RSP_REASON_BLOCKED:
1472 case RSP_REASON_UNTRUSTED:
1473 reason = "Untrusted";
1475 case RSP_REASON_REDIRECTED:
1476 reason = "Redirected";
1478 case RSP_REASON_CGI_CALL:
1479 reason = "CGI Call";
1481 case RSP_REASON_NO_SUCH_DOMAIN:
1482 reason = "DNS failure";
1484 case RSP_REASON_FORWARDING_FAILED:
1485 reason = "Forwarding failed";
1487 case RSP_REASON_CONNECT_FAILED:
1488 reason = "Connection failure";
1490 case RSP_REASON_OUT_OF_MEMORY:
1491 reason = "Out of memory (may mask other reasons)";
1494 reason = "No reason recorded";
1502 /*********************************************************************
1504 * Function : send_crunch_response
1506 * Description : Delivers already prepared response for
1507 * intercepted requests, logs the interception
1508 * and frees the response.
1511 * 1 : csp = Current client state (buffers, headers, etc...)
1512 * 1 : rsp = Fully prepared response. Will be freed on exit.
1514 * Returns : Nothing.
1516 *********************************************************************/
1517 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1519 const struct http_request *http = csp->http;
1520 char status_code[4];
1522 assert(rsp != NULL);
1523 assert(rsp->head != NULL);
1528 * Not supposed to happen. If it does
1529 * anyway, treat it as an unknown error.
1531 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1532 /* return code doesn't matter */
1537 /* If rsp is still NULL, we have serious internal problems. */
1538 log_error(LOG_LEVEL_FATAL,
1539 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1543 * Extract the status code from the actual head
1544 * that was send to the client. It is the only
1545 * way to get it right for all requests, including
1546 * the fixed ones for out-of-memory problems.
1548 * A head starts like this: 'HTTP/1.1 200...'
1552 status_code[0] = rsp->head[9];
1553 status_code[1] = rsp->head[10];
1554 status_code[2] = rsp->head[11];
1555 status_code[3] = '\0';
1557 /* Write the answer to the client */
1558 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1559 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1561 /* There is nothing we can do about it. */
1562 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1565 /* Log that the request was crunched and why. */
1566 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1567 http->hostport, http->path, crunch_reason(rsp));
1568 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1569 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1571 /* Clean up and return */
1572 if (cgi_error_memory() != rsp)
1574 free_http_response(rsp);
1581 /*********************************************************************
1583 * Function : request_contains_null_bytes
1585 * Description : Checks for NULL bytes in the request and sends
1586 * an error message to the client if any were found.
1588 * XXX: currently not used, see comment in chat().
1591 * 1 : csp = Current client state (buffers, headers, etc...)
1592 * 2 : buf = Data from the client's request to check.
1593 * 3 : len = The data length.
1595 * Returns : TRUE if the request contained one or more NULL bytes, or
1598 *********************************************************************/
1599 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1601 size_t c_len; /* Request lenght when treated as C string */
1603 c_len = strlen(buf);
1608 * Null byte(s) found. Log the request,
1609 * return an error response and hang up.
1611 size_t tmp_len = c_len;
1616 * Replace NULL byte(s) with '°' characters
1617 * so the request can be logged as string.
1618 * XXX: Is there a better replacement character?
1621 tmp_len += strlen(buf+tmp_len);
1622 } while (tmp_len < len);
1624 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1625 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1626 log_error(LOG_LEVEL_HEADER,
1627 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1629 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1631 /* XXX: Log correct size */
1632 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1642 /*********************************************************************
1644 * Function : crunch_response_triggered
1646 * Description : Checks if the request has to be crunched,
1647 * and delivers the crunch response if necessary.
1650 * 1 : csp = Current client state (buffers, headers, etc...)
1651 * 2 : crunchers = list of cruncher functions to run
1653 * Returns : TRUE if the request was answered with a crunch response
1656 *********************************************************************/
1657 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1659 struct http_response *rsp = NULL;
1660 const struct cruncher *c;
1663 * If CGI request crunching is disabled,
1664 * check the CGI dispatcher out of order to
1665 * prevent unintentional blocks or redirects.
1667 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1668 && (NULL != (rsp = dispatch_cgi(csp))))
1670 /* Deliver, log and free the interception response. */
1671 send_crunch_response(csp, rsp);
1675 for (c = crunchers; c->cruncher != NULL; c++)
1678 * Check the cruncher if either Privoxy is toggled
1679 * on and the request isn't forced, or if the cruncher
1680 * applies to forced requests as well.
1682 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1683 !(csp->flags & CSP_FLAG_FORCED)) ||
1684 (c->flags & CF_IGNORE_FORCE))
1686 rsp = c->cruncher(csp);
1689 /* Deliver, log and free the interception response. */
1690 send_crunch_response(csp, rsp);
1691 #ifdef FEATURE_STATISTICS
1692 if (c->flags & CF_COUNT_AS_REJECT)
1694 csp->flags |= CSP_FLAG_REJECTED;
1696 #endif /* def FEATURE_STATISTICS */
1707 /*********************************************************************
1709 * Function : build_request_line
1711 * Description : Builds the HTTP request line.
1713 * If a HTTP forwarder is used it expects the whole URL,
1714 * web servers only get the path.
1717 * 1 : csp = Current client state (buffers, headers, etc...)
1718 * 2 : fwd = The forwarding spec used for the request
1719 * XXX: Should use http->fwd instead.
1720 * 3 : request_line = The old request line which will be replaced.
1722 * Returns : Nothing. Terminates in case of memory problems.
1724 *********************************************************************/
1725 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1727 struct http_request *http = csp->http;
1729 assert(http->ssl == 0);
1732 * Downgrade http version from 1.1 to 1.0
1733 * if +downgrade action applies.
1735 if ( (csp->action->flags & ACTION_DOWNGRADE)
1736 && (!strcmpic(http->ver, "HTTP/1.1")))
1739 http->ver = strdup("HTTP/1.0");
1741 if (http->ver == NULL)
1743 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1748 * Rebuild the request line.
1750 freez(*request_line);
1751 *request_line = strdup(http->gpc);
1752 string_append(request_line, " ");
1754 if (fwd->forward_host)
1756 string_append(request_line, http->url);
1760 string_append(request_line, http->path);
1762 string_append(request_line, " ");
1763 string_append(request_line, http->ver);
1765 if (*request_line == NULL)
1767 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1769 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1773 /*********************************************************************
1775 * Function : change_request_destination
1777 * Description : Parse a (rewritten) request line and regenerate
1778 * the http request data.
1781 * 1 : csp = Current client state (buffers, headers, etc...)
1783 * Returns : Forwards the parse_http_request() return code.
1784 * Terminates in case of memory problems.
1786 *********************************************************************/
1787 static jb_err change_request_destination(struct client_state *csp)
1789 struct http_request *http = csp->http;
1792 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1793 free_http_request(http);
1794 err = parse_http_request(csp->headers->first->str, http, csp);
1795 if (JB_ERR_OK != err)
1797 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1798 jb_err_to_string(err));
1800 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1801 if (http->ocmd == NULL)
1803 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1810 /*********************************************************************
1812 * Function : get_filter_function
1814 * Description : Decides which content filter function has
1815 * to be applied (if any).
1817 * XXX: Doesn't handle filter_popups()
1818 * because of the different prototype. Probably
1819 * we should ditch filter_popups() anyway, it's
1820 * even less reliable than popup blocking based
1824 * 1 : csp = Current client state (buffers, headers, etc...)
1826 * Returns : The content filter function to run, or
1827 * NULL if no content filter is active
1829 *********************************************************************/
1830 static filter_function_ptr get_filter_function(struct client_state *csp)
1832 filter_function_ptr filter_function = NULL;
1835 * Are we enabling text mode by force?
1837 if (csp->action->flags & ACTION_FORCE_TEXT_MODE)
1840 * Do we really have to?
1842 if (csp->content_type & CT_TEXT)
1844 log_error(LOG_LEVEL_HEADER, "Text mode is already enabled.");
1848 csp->content_type |= CT_TEXT;
1849 log_error(LOG_LEVEL_HEADER, "Text mode enabled by force. Take cover!");
1853 if (!(csp->content_type & CT_DECLARED))
1856 * The server didn't bother to declare a MIME-Type.
1857 * Assume it's text that can be filtered.
1859 * This also regulary happens with 304 responses,
1860 * therefore logging anything here would cause
1863 csp->content_type |= CT_TEXT;
1868 * Choose the applying filter function based on
1869 * the content type and action settings.
1871 if ((csp->content_type & CT_TEXT) &&
1872 (csp->rlist != NULL) &&
1873 (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER])))
1875 filter_function = pcrs_filter_response;
1877 else if ((csp->content_type & CT_GIF) &&
1878 (csp->action->flags & ACTION_DEANIMATE))
1880 filter_function = gif_deanimate_response;
1882 else if ((csp->content_type & CT_JPEG) &&
1883 (csp->action->flags & ACTION_JPEG_INSPECT))
1885 filter_function = jpeg_inspect_response;
1888 return filter_function;
1891 /*********************************************************************
1895 * Description : Once a connection to the client has been accepted,
1896 * this function is called (via serve()) to handle the
1897 * main business of the communication. When this
1898 * function returns, the caller must close the client
1901 * FIXME: chat is nearly thousand lines long.
1905 * 1 : csp = Current client state (buffers, headers, etc...)
1907 * Returns : Nothing.
1909 *********************************************************************/
1910 static void chat(struct client_state *csp)
1912 char buf[BUFFER_SIZE];
1920 int ms_iis5_hack = 0;
1921 size_t byte_count = 0;
1922 int forwarded_connect_retries = 0;
1923 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1924 const struct forward_spec * fwd;
1925 struct http_request *http;
1926 int len; /* for buffer sizes (and negative error codes) */
1928 #ifdef FEATURE_KILL_POPUPS
1929 int block_popups_now = 0; /* bool, 1==currently blocking popups */
1930 #endif /* def FEATURE_KILL_POPUPS */
1932 /* Function that does the content filtering for the current request */
1933 filter_function_ptr content_filter = NULL;
1935 /* Skeleton for HTTP response, if we should intercept the request */
1936 struct http_response *rsp;
1938 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1939 struct list header_list;
1940 struct list *headers = &header_list;
1944 memset(buf, 0, sizeof(buf));
1947 * Read the client's request. Note that since we're not using select() we
1948 * could get blocked here if a client connected, then didn't say anything!
1953 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1955 if (len <= 0) break; /* error! */
1958 * If there is no memory left for buffering the
1959 * request, there is nothing we can do but hang up
1961 if (add_to_iob(csp, buf, len))
1966 req = get_header(csp);
1968 } while ((NULL != req) && ('\0' == *req));
1972 /* Request received. Validate and parse it. */
1976 * XXX: Temporary disabled to prevent problems
1977 * with POST requests whose bodies are allowed to
1978 * contain NULL bytes. BR#1730105.
1980 * The main purpose of this check is to properly
1981 * log stuff like BitTorrent traffic and other junk
1982 * that hits public proxies. It's not required for
1983 * Privoxy to functions as those requests are discarded
1986 * It probably should be rewritten to only check
1987 * the head of the request. Another option would
1988 * be to let all POST requests pass, although that
1989 * may not be good enough.
1991 if (request_contains_null_bytes(csp, buf, len))
1993 /* NULL bytes found and dealt with, just hang up. */
1998 /* Does the request line look invalid? */
1999 if (client_protocol_is_unsupported(csp, req))
2002 * Yes. The request has already been
2003 * answered with a error response, the buffers
2004 * were freed and we're done with chatting.
2009 #ifdef FEATURE_FORCE_LOAD
2011 * If this request contains the FORCE_PREFIX and blocks
2012 * aren't enforced, get rid of it and set the force flag.
2014 if (strstr(req, FORCE_PREFIX))
2016 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2018 log_error(LOG_LEVEL_FORCE,
2019 "Ignored force prefix in request: \"%s\".", req);
2023 strclean(req, FORCE_PREFIX);
2024 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2025 csp->flags |= CSP_FLAG_FORCED;
2029 #endif /* def FEATURE_FORCE_LOAD */
2030 err = parse_http_request(req, http, csp);
2031 if (JB_ERR_OK != err)
2033 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2039 if (http->cmd == NULL)
2041 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2042 /* XXX: Use correct size */
2043 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2044 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2046 free_http_request(http);
2050 /* grab the rest of the client's headers */
2054 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
2056 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2059 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2064 * If there is no memory left for buffering the
2065 * request, there is nothing we can do but hang up
2067 if (add_to_iob(csp, buf, len))
2074 if (p == NULL) break;
2081 if (http->host == NULL)
2084 * If we still don't know the request destination,
2085 * the request is invalid or the client uses
2086 * Privoxy without its knowledge.
2088 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2091 * Our attempts to get the request destination
2092 * elsewhere failed or Privoxy is configured
2093 * to only accept proxy requests.
2095 * An error response has already been send
2096 * and we're done here.
2103 * Determine the actions for this URL
2105 #ifdef FEATURE_TOGGLE
2106 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2108 /* Most compatible set of actions (i.e. none) */
2109 init_current_action(csp->action);
2112 #endif /* ndef FEATURE_TOGGLE */
2114 url_actions(http, csp);
2118 * Save a copy of the original request for logging
2120 http->ocmd = strdup(http->cmd);
2122 if (http->ocmd == NULL)
2124 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2127 enlist(csp->headers, http->cmd);
2129 /* Append the previously read headers */
2130 list_append_list_unique(csp->headers, headers);
2131 destroy_list(headers);
2134 * If the user has not supplied any wafers, and the user has not
2135 * told us to suppress the vanilla wafer, then send the vanilla wafer.
2137 if (list_is_empty(csp->action->multi[ACTION_MULTI_WAFER])
2138 && ((csp->action->flags & ACTION_VANILLA_WAFER) != 0))
2140 enlist(csp->action->multi[ACTION_MULTI_WAFER], VANILLA_WAFER);
2143 err = sed(client_patterns, add_client_headers, csp);
2144 if (JB_ERR_OK != err)
2146 assert(err == JB_ERR_PARSE);
2147 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2149 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2151 /* Check request line for rewrites. */
2152 if ((NULL == csp->headers->first->str)
2153 || (strcmp(http->cmd, csp->headers->first->str) &&
2154 (JB_ERR_OK != change_request_destination(csp))))
2157 * A header filter broke the request line - bail out.
2159 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2160 /* XXX: Use correct size */
2161 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2162 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2164 free_http_request(http);
2168 /* decide how to route the HTTP request */
2169 if (NULL == (fwd = forward_url(http, csp)))
2171 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2172 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2175 /* build the http request to send to the server
2176 * we have to do one of the following:
2178 * create = use the original HTTP request to create a new
2179 * HTTP request that has either the path component
2180 * without the http://domainspec (w/path) or the
2181 * full orininal URL (w/url)
2182 * Note that the path and/or the HTTP version may
2183 * have been altered by now.
2185 * connect = Open a socket to the host:port of the server
2186 * and short-circuit server and client socket.
2188 * pass = Pass the request unchanged if forwarding a CONNECT
2189 * request to a parent proxy. Note that we'll be sending
2190 * the CFAIL message ourselves if connecting to the parent
2191 * fails, but we won't send a CSUCCEED message if it works,
2192 * since that would result in a double message (ours and the
2193 * parent's). After sending the request to the parent, we simply
2196 * here's the matrix:
2199 * +--------+--------+
2201 * 0 | create | connect|
2203 * Forwarding +--------+--------+
2205 * 1 | create | pass |
2207 * +--------+--------+
2212 * Check if a CONNECT request is allowable:
2213 * In the absence of a +limit-connect action, allow only port 443.
2214 * If there is an action, allow whatever matches the specificaton.
2218 if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443)
2219 || (csp->action->flags & ACTION_LIMIT_CONNECT
2220 && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) )
2222 if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS)
2225 * The response may confuse some clients,
2226 * but makes unblocking easier.
2228 * XXX: It seems to work with all major browsers,
2229 * so we should consider returning a body by default someday ...
2231 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2232 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2233 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2235 csp->action->flags |= ACTION_BLOCK;
2240 write_socket(csp->cfd, CFORBIDDEN, strlen(CFORBIDDEN));
2241 log_error(LOG_LEVEL_INFO, "Request from %s denied. "
2242 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2243 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2245 assert(NULL != csp->http->ocmd);
2246 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 403 0", csp->ip_addr_str, csp->http->ocmd);
2248 list_remove_all(csp->headers);
2250 * XXX: For consistency we might want to log a crunch message here.
2259 freez(csp->headers->first->str);
2260 build_request_line(csp, fwd, &csp->headers->first->str);
2263 hdr = list_to_text(csp->headers);
2266 /* FIXME Should handle error properly */
2267 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2271 * We have a request. Check if one of the crunchers wants it.
2273 if (crunch_response_triggered(csp, crunchers_all))
2276 * Yes. The client got the crunch response
2277 * and we are done here after cleaning up.
2280 list_remove_all(csp->headers);
2286 * The headers can't be removed earlier because
2287 * they were still needed for the referrer check
2288 * in case of CGI crunches.
2290 * XXX: Would it be worth to move the referrer check
2291 * into client_referrer() and set a flag if it's trusted?
2293 list_remove_all(csp->headers);
2295 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2297 if (fwd->forward_host)
2299 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2300 fwd->forward_host, fwd->forward_port, http->hostport);
2304 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2307 /* here we connect to the server, gateway, or the forwarder */
2309 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2310 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2312 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2313 forwarded_connect_retries, http->hostport);
2316 if (csp->sfd == JB_INVALID_SOCKET)
2318 if (fwd->type != SOCKS_NONE)
2321 rsp = error_response(csp, "forwarding-failed", errno);
2323 else if (errno == EINVAL)
2325 rsp = error_response(csp, "no-such-domain", errno);
2329 rsp = error_response(csp, "connect-failed", errno);
2330 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2335 /* Write the answer to the client */
2338 send_crunch_response(csp, rsp);
2345 if (fwd->forward_host || (http->ssl == 0))
2347 /* write the client's (modified) header to the server
2348 * (along with anything else that may be in the buffer)
2351 if (write_socket(csp->sfd, hdr, strlen(hdr))
2352 || (flush_socket(csp->sfd, csp) < 0))
2354 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2357 rsp = error_response(csp, "connect-failed", errno);
2361 send_crunch_response(csp, rsp);
2371 * We're running an SSL tunnel and we're not forwarding,
2372 * so just send the "connect succeeded" message to the
2373 * client, flush the rest, and get out of the way.
2375 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2383 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2385 /* we're finished with the client's header */
2388 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2390 /* pass data between the client and server
2391 * until one or the other shuts down the connection.
2400 * FD_ZERO here seems to point to an errant macro which crashes.
2401 * So do this by hand for now...
2403 memset(&rfds,0x00,sizeof(fd_set));
2407 FD_SET(csp->cfd, &rfds);
2408 FD_SET(csp->sfd, &rfds);
2410 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2414 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2418 /* this is the body of the browser's request
2419 * just read it and write it.
2422 if (FD_ISSET(csp->cfd, &rfds))
2424 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2428 break; /* "game over, man" */
2431 if (write_socket(csp->sfd, buf, (size_t)len))
2433 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2440 * The server wants to talk. It could be the header or the body.
2441 * If `hdr' is null, then it's the header otherwise it's the body.
2442 * FIXME: Does `hdr' really mean `host'? No.
2446 if (FD_ISSET(csp->sfd, &rfds))
2449 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2453 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2455 if (http->ssl && (fwd->forward_host == NULL))
2458 * Just hang up. We already confirmed the client's CONNECT
2459 * request with status code 200 and unencrypted content is
2460 * no longer welcome.
2462 log_error(LOG_LEVEL_ERROR,
2463 "CONNECT already confirmed. Unable to tell the client about the problem.");
2466 else if (byte_count)
2469 * Just hang up. We already transmitted the original headers
2470 * and parts of the original content and therefore missed the
2471 * chance to send an error message (without risking data corruption).
2473 * XXX: we could retry with a fancy range request here.
2475 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2476 "Unable to tell the client about the problem.");
2480 rsp = error_response(csp, "connect-failed", errno);
2484 send_crunch_response(csp, rsp);
2490 /* Add a trailing zero. This lets filter_popups
2491 * use string operations.
2495 #ifdef FEATURE_KILL_POPUPS
2496 /* Filter the popups on this read. */
2497 if (block_popups_now)
2499 filter_popups(buf, csp);
2501 #endif /* def FEATURE_KILL_POPUPS */
2503 /* Normally, this would indicate that we've read
2504 * as much as the server has sent us and we can
2505 * close the client connection. However, Microsoft
2506 * in its wisdom has released IIS/5 with a bug that
2507 * prevents it from sending the trailing \r\n in
2508 * a 302 redirect header (and possibly other headers).
2509 * To work around this if we've haven't parsed
2510 * a full header we'll append a trailing \r\n
2511 * and see if this now generates a valid one.
2513 * This hack shouldn't have any impacts. If we've
2514 * already transmitted the header or if this is a
2515 * SSL connection, then we won't bother with this
2516 * hack. So we only work on partially received
2517 * headers. If we append a \r\n and this still
2518 * doesn't generate a valid header, then we won't
2519 * transmit anything to the client.
2524 if (server_body || http->ssl)
2527 * If we have been buffering up the document,
2528 * now is the time to apply content modification
2529 * and send the result to the client.
2534 * If the content filter fails, use the original
2535 * buffer and length.
2536 * (see p != NULL ? p : csp->iob->cur below)
2538 if (NULL == (p = (*content_filter)(csp)))
2540 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2543 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2545 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2548 hdr = list_to_text(csp->headers);
2551 /* FIXME Should handle error properly */
2552 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2556 * Shouldn't happen because this was the second sed run
2557 * and tags are only created for the first one.
2559 assert(!crunch_response_triggered(csp, crunchers_all));
2561 if (write_socket(csp->cfd, hdr, strlen(hdr))
2562 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2564 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2574 break; /* "game over, man" */
2578 * This is NOT the body, so
2579 * Let's pretend the server just sent us a blank line.
2581 snprintf(buf, sizeof(buf), "\r\n");
2582 len = (int)strlen(buf);
2585 * Now, let the normal header parsing algorithm below do its
2586 * job. If it fails, we'll exit instead of continuing.
2593 * If this is an SSL connection or we're in the body
2594 * of the server document, just write it to the client,
2595 * unless we need to buffer the body for later content-filtering
2598 if (server_body || http->ssl)
2603 * If there is no memory left for buffering the content, or the buffer limit
2604 * has been reached, switch to non-filtering mode, i.e. make & write the
2605 * header, flush the iob and buf, and get out of the way.
2607 if (add_to_iob(csp, buf, len))
2612 log_error(LOG_LEVEL_ERROR, "Flushing header and buffers. Stepping back from filtering.");
2614 hdr = list_to_text(csp->headers);
2618 * Memory is too tight to even generate the header.
2619 * Send our static "Out-of-memory" page.
2621 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2622 rsp = cgi_error_memory();
2623 send_crunch_response(csp, rsp);
2627 hdrlen = strlen(hdr);
2629 if (write_socket(csp->cfd, hdr, hdrlen)
2630 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
2631 || (write_socket(csp->cfd, buf, (size_t)len)))
2633 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2640 * Reset the byte_count to the amount of bytes
2641 * we just flushed. len will be added a few lines below,
2642 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2644 byte_count = (size_t)flushed;
2646 content_filter = NULL;
2652 if (write_socket(csp->cfd, buf, (size_t)len))
2654 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2658 byte_count += (size_t)len;
2663 /* we're still looking for the end of the
2664 * server's header ... (does that make header
2665 * parsing an "out of body experience" ?
2669 * buffer up the data we just read. If that fails,
2670 * there's little we can do but send our static
2671 * out-of-memory page.
2673 if (add_to_iob(csp, buf, len))
2675 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2676 rsp = cgi_error_memory();
2677 send_crunch_response(csp, rsp);
2682 /* Convert iob into something sed() can digest */
2683 if (JB_ERR_PARSE == get_server_headers(csp))
2687 /* Well, we tried our MS IIS/5
2688 * hack and it didn't work.
2689 * The header is incomplete
2690 * and there isn't anything
2691 * we can do about it.
2697 /* Since we have to wait for
2698 * more from the server before
2699 * we can parse the headers
2700 * we just continue here.
2706 /* Did we actually get anything? */
2707 if (NULL == csp->headers->first)
2709 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2710 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2711 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2712 free_http_request(http);
2716 /* we have now received the entire header.
2717 * filter it and send the result to the client
2719 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2721 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2723 hdr = list_to_text(csp->headers);
2726 /* FIXME Should handle error properly */
2727 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2730 if (crunch_response_triggered(csp, crunchers_light))
2733 * One of the tags created by a server-header
2734 * tagger triggered a crunch. We already
2735 * delivered the crunch response to the client
2736 * and are done here after cleaning up.
2741 /* Buffer and pcrs filter this if appropriate. */
2743 if (!http->ssl) /* We talk plaintext */
2746 #ifdef FEATURE_KILL_POPUPS
2747 /* Start blocking popups if appropriate. */
2748 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2749 (csp->action->flags & ACTION_NO_POPUPS) != 0) /* Policy allows */
2751 block_popups_now = 1;
2753 * Filter the part of the body that came in the same read
2754 * as the last headers:
2756 filter_popups(csp->iob->cur, csp);
2758 #endif /* def FEATURE_KILL_POPUPS */
2759 content_filter = get_filter_function(csp);
2762 * Only write if we're not buffering for content modification
2764 if (!content_filter)
2766 /* write the server's (modified) header to
2767 * the client (along with anything else that
2768 * may be in the buffer)
2771 if (write_socket(csp->cfd, hdr, strlen(hdr))
2772 || ((len = flush_socket(csp->cfd, csp)) < 0))
2774 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2776 /* the write failed, so don't bother
2777 * mentioning it to the client...
2778 * it probably can't hear us anyway.
2784 byte_count += (size_t)len;
2787 /* we're finished with the server's header */
2792 /* If this was a MS IIS/5 hack then it means
2793 * the server has already closed the
2794 * connection. Nothing more to read. Time
2805 return; /* huh? we should never get here */
2808 if (csp->content_length == 0)
2811 * If Privoxy didn't recalculate the
2812 * Content-Lenght, byte_count is still
2815 csp->content_length = byte_count;
2818 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2819 csp->ip_addr_str, http->ocmd, csp->content_length);
2823 /*********************************************************************
2827 * Description : This is little more than chat. We only "serve" to
2828 * to close any socket that chat may have opened.
2831 * 1 : csp = Current client state (buffers, headers, etc...)
2835 *********************************************************************/
2837 void serve(struct client_state *csp)
2838 #else /* ifndef AMIGA */
2839 static void serve(struct client_state *csp)
2840 #endif /* def AMIGA */
2843 close_socket(csp->cfd);
2845 if (csp->sfd != JB_INVALID_SOCKET)
2847 close_socket(csp->sfd);
2850 csp->flags &= ~CSP_FLAG_ACTIVE;
2856 /*********************************************************************
2858 * Function : server_thread
2860 * Description : We only exist to call `serve' in a threaded environment.
2863 * 1 : data = Current client state (buffers, headers, etc...)
2865 * Returns : Always 0.
2867 *********************************************************************/
2868 static int32 server_thread(void *data)
2870 serve((struct client_state *) data);
2878 /*********************************************************************
2882 * Description : Print usage info & exit.
2884 * Parameters : Pointer to argv[0] for identifying ourselves
2888 *********************************************************************/
2889 static void usage(const char *myname)
2891 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2895 #endif /* defined(unix) */
2898 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2899 #endif /* defined(unix) */
2900 "[--version] [configfile]\n"
2901 "Aborting\n", myname);
2906 #endif /* defined(unix) */
2909 /*********************************************************************
2911 * Function : initialize_mutexes
2913 * Description : Prepares mutexes if mutex support is available.
2917 * Returns : Void, exits in case of errors.
2919 *********************************************************************/
2920 static void initialize_mutexes(void)
2924 #ifdef FEATURE_PTHREAD
2926 * Prepare global mutex semaphores
2928 err = pthread_mutex_init(&log_mutex, 0);
2930 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2933 * XXX: The assumptions below are a bit naive
2934 * and can cause locks that aren't necessary.
2936 * For example older FreeBSD versions (< 6.x?)
2937 * have no gethostbyname_r, but gethostbyname is
2940 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2941 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2942 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2944 * XXX: should we use a single mutex for
2945 * localtime() and gmtime() as well?
2947 #ifndef HAVE_GMTIME_R
2948 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2949 #endif /* ndef HAVE_GMTIME_R */
2951 #ifndef HAVE_LOCALTIME_R
2952 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2953 #endif /* ndef HAVE_GMTIME_R */
2956 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2957 #endif /* ndef HAVE_RANDOM */
2958 #endif /* FEATURE_PTHREAD */
2961 * TODO: mutex support for mingw32 would be swell.
2966 printf("Fatal error. Mutex initialization failed: %s.\n",
2975 /*********************************************************************
2979 * Description : Load the config file and start the listen loop.
2980 * This function is a lot more *sane* with the `load_config'
2981 * and `listen_loop' functions; although it stills does
2982 * a *little* too much for my taste.
2985 * 1 : argc = Number of parameters (including $0).
2986 * 2 : argv = Array of (char *)'s to the parameters.
2988 * Returns : 1 if : can't open config file, unrecognized directive,
2989 * stats requested in multi-thread mode, can't open the
2990 * log file, can't open the jar file, listen port is invalid,
2991 * any load fails, and can't bind port.
2993 * Else main never returns, the process must be signaled
2994 * to terminate execution. Or, on Windows, use the
2995 * "File", "Exit" menu option.
2997 *********************************************************************/
2999 int real_main(int argc, const char *argv[])
3001 int main(int argc, const char *argv[])
3005 unsigned int random_seed;
3007 struct passwd *pw = NULL;
3008 struct group *grp = NULL;
3011 char *pre_chroot_nslookup_to_load_resolver = NULL;
3018 #if !defined(_WIN32)
3026 * Parse the command line arguments
3028 * XXX: simply printing usage information in case of
3029 * invalid arguments isn't particular user friendly.
3031 while (++argc_pos < argc)
3034 /* Check to see if the service must be installed or uninstalled */
3035 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3037 const char *pName = argv[argc_pos] + 9;
3040 exit( (install_service(pName)) ? 0 : 1 );
3042 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3044 const char *pName = argv[argc_pos] + 11;
3047 exit((uninstall_service(pName)) ? 0 : 1);
3049 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3051 bRunAsService = TRUE;
3052 w32_set_service_cwd();
3053 atexit(w32_service_exit_notify);
3056 #endif /* defined(_WIN32) */
3059 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3061 if (strcmp(argv[argc_pos], "--help") == 0)
3066 else if(strcmp(argv[argc_pos], "--version") == 0)
3068 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3074 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3079 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3081 if (++argc_pos == argc) usage(argv[0]);
3082 pidfile = strdup(argv[argc_pos]);
3085 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3087 if (++argc_pos == argc) usage(argv[argc_pos]);
3089 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3092 if (NULL == (grp = getgrnam(p)))
3094 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3098 if (NULL == (pw = getpwnam(argv[argc_pos])))
3100 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3103 if (p != NULL) *--p = '\0';
3106 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3108 if (++argc_pos == argc) usage(argv[0]);
3109 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3112 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3116 #endif /* defined(unix) */
3118 else if (argc_pos + 1 != argc)
3121 * This is neither the last command line
3122 * option, nor was it recognized before,
3123 * therefore it must be invalid.
3129 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3131 configfile = argv[argc_pos];
3134 } /* -END- while (more arguments) */
3137 if ( *configfile != '/' )
3139 char cwd[BUFFER_SIZE];
3141 size_t abs_file_size;
3143 /* make config-filename absolute here */
3144 if (NULL == getcwd(cwd, sizeof(cwd)))
3146 perror("failed to get current working directory");
3151 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3152 basedir = strdup(cwd);
3154 if (NULL == basedir ||
3155 NULL == (abs_file = malloc(abs_file_size)))
3157 perror("malloc failed");
3160 strlcpy(abs_file, basedir, abs_file_size);
3161 strlcat(abs_file, "/", abs_file_size );
3162 strlcat(abs_file, configfile, abs_file_size);
3163 configfile = abs_file;
3165 #endif /* defined unix */
3169 clients->next = NULL;
3173 #elif defined(_WIN32)
3177 /* Prepare mutexes if supported and necessary. */
3178 initialize_mutexes();
3180 random_seed = (unsigned int)time(NULL);
3182 srandom(random_seed);
3185 #endif /* ifdef HAVE_RANDOM */
3188 * Unix signal handling
3190 * Catch the abort, interrupt and terminate signals for a graceful exit
3191 * Catch the hangup signal so the errlog can be reopened.
3192 * Ignore the broken pipe signals (FIXME: Why?)
3194 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3197 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3198 const int ignored_signals[] = { SIGPIPE, 0 };
3200 for (idx = 0; catched_signals[idx] != 0; idx++)
3202 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3203 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3205 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3206 #endif /* ifdef sun */
3208 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3212 for (idx = 0; ignored_signals[idx] != 0; idx++)
3214 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3216 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3221 #else /* ifdef _WIN32 */
3222 # ifdef _WIN_CONSOLE
3224 * We *are* in a windows console app.
3225 * Print a verbose messages about FAQ's and such
3227 printf("%s", win32_blurb);
3228 # endif /* def _WIN_CONSOLE */
3229 #endif /* def _WIN32 */
3232 /* Initialize the CGI subsystem */
3233 cgi_init_error_messages();
3236 * If runnig on unix and without the --nodaemon
3237 * option, become a daemon. I.e. fork, detach
3238 * from tty and get process group leadership
3251 if ( pid < 0 ) /* error */
3256 else if ( pid != 0 ) /* parent */
3261 * must check for errors
3262 * child died due to missing files aso
3265 wpid = waitpid( pid, &status, WNOHANG );
3274 /* Should be more portable, but not as well tested */
3279 #else /* ndef __FreeBSD__ */
3281 #endif /* ndef __FreeBSD__ */
3282 fd = open("/dev/tty", O_RDONLY);
3285 /* no error check here */
3286 ioctl( fd, TIOCNOTTY,0 );
3290 /* FIXME: should close stderr (fd 2) here too, but the test
3292 * and load config file is done in listen_loop() and puts
3293 * some messages on stderr there.
3300 } /* -END- if (!no_daemon) */
3303 * As soon as we have written the PID file, we can switch
3304 * to the user and group ID indicated by the --user option
3310 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3312 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3318 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3320 /* Read the time zone file from /etc before doing chroot. */
3322 if (NULL != pre_chroot_nslookup_to_load_resolver
3323 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3325 /* Initialize resolver library. */
3326 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3328 if (chroot(pw->pw_dir) < 0)
3330 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3334 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3337 if (setuid(pw->pw_uid))
3339 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3343 char putenv_dummy[64];
3345 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3346 if (putenv(putenv_dummy) != 0)
3348 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3351 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3352 if (putenv(putenv_dummy) != 0)
3354 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3360 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3363 #endif /* defined unix */
3366 /* This will be FALSE unless the command line specified --service
3370 /* Yup, so now we must attempt to establish a connection
3371 * with the service dispatcher. This will only work if this
3372 * process was launched by the service control manager to
3373 * actually run as a service. If this isn't the case, i've
3374 * known it take around 30 seconds or so for the call to return.
3377 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3378 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3380 /* Service has run, and at this point is now being stopped, so just return */
3385 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3387 /* An error occurred. Usually it's because --service was wrongly specified
3388 * and we were unable to connect to the Service Control Dispatcher because
3389 * it wasn't expecting us and is therefore not listening.
3391 * For now, just continue below to call the listen_loop function.
3394 #endif /* def _WIN32 */
3404 /*********************************************************************
3406 * Function : bind_port_helper
3408 * Description : Bind the listen port. Handles logging, and aborts
3412 * 1 : config = Privoxy configuration. Specifies port
3415 * Returns : Port that was opened.
3417 *********************************************************************/
3418 static jb_socket bind_port_helper(struct configuration_spec * config)
3423 if ( (config->haddr != NULL)
3424 && (config->haddr[0] == '1')
3425 && (config->haddr[1] == '2')
3426 && (config->haddr[2] == '7')
3427 && (config->haddr[3] == '.') )
3429 log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only",
3432 else if (config->haddr == NULL)
3434 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3439 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3440 config->hport, config->haddr);
3443 result = bind_port(config->haddr, config->hport, &bfd);
3450 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3451 "There may be another Privoxy or some other "
3452 "proxy running on port %d",
3453 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3454 config->hport, config->hport);
3457 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3458 "The hostname is not resolvable",
3459 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3462 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3463 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3466 /* shouldn't get here */
3467 return JB_INVALID_SOCKET;
3470 config->need_bind = 0;
3477 /* Without this simple workaround we get this compiler warning from _beginthread
3478 * warning C4028: formal parameter 1 different from declaration
3480 void w32_service_listen_loop(void *p)
3484 #endif /* def _WIN32 */
3487 /*********************************************************************
3489 * Function : listen_loop
3491 * Description : bind the listen port and enter a "FOREVER" listening loop.
3497 *********************************************************************/
3498 static void listen_loop(void)
3500 struct client_state *csp = NULL;
3502 struct configuration_spec * config;
3504 config = load_config();
3506 bfd = bind_port_helper(config);
3508 #ifdef FEATURE_GRACEFUL_TERMINATION
3509 while (!g_terminate)
3514 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3515 while (waitpid(-1, NULL, WNOHANG) > 0)
3517 /* zombie children */
3519 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3522 * Free data that was used by died threads
3528 * Re-open the errlog after HUP signal
3530 if (received_hup_signal)
3532 init_error_log(Argv[0], config->logfile, config->debug);
3533 received_hup_signal = 0;
3538 #ifdef FEATURE_COOKIE_JAR
3540 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3541 * already open. I think unload_configfile() is not being run, which should do
3542 * this work. Until that can get resolved, we'll use this workaround.
3546 if (csp->config->jar)
3548 fclose(csp->config->jar);
3549 csp->config->jar = NULL;
3551 #endif /* FEATURE_COOKIE_JAR */
3552 #endif /* __OS2__ */
3554 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3556 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3560 csp->flags |= CSP_FLAG_ACTIVE;
3561 csp->sfd = JB_INVALID_SOCKET;
3563 csp->config = config = load_config();
3565 if ( config->need_bind )
3568 * Since we were listening to the "old port", we will not see
3569 * a "listen" param change until the next IJB request. So, at
3570 * least 1 more request must be made for us to find the new
3571 * setting. I am simply closing the old socket and binding the
3574 * Which-ever is correct, we will serve 1 more page via the
3575 * old settings. This should probably be a "show-proxy-args"
3576 * request. This should not be a so common of an operation
3577 * that this will hurt people's feelings.
3582 bfd = bind_port_helper(config);
3585 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3587 if (!accept_connection(csp, bfd))
3589 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3602 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3605 #ifdef FEATURE_TOGGLE
3606 if (global_toggle_state)
3608 csp->flags |= CSP_FLAG_TOGGLED_ON;
3610 #endif /* def FEATURE_TOGGLE */
3612 if (run_loader(csp))
3614 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3615 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3619 if (block_acl(NULL,csp))
3621 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3622 close_socket(csp->cfd);
3626 #endif /* def FEATURE_ACL */
3628 /* add it to the list of clients */
3629 csp->next = clients->next;
3630 clients->next = csp;
3632 if (config->multi_threaded)
3636 /* this is a switch () statment in the C preprocessor - ugh */
3637 #undef SELECTED_ONE_OPTION
3639 /* Use Pthreads in preference to native code */
3640 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3641 #define SELECTED_ONE_OPTION
3643 pthread_t the_thread;
3644 pthread_attr_t attrs;
3646 pthread_attr_init(&attrs);
3647 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3648 errno = pthread_create(&the_thread, &attrs,
3649 (void * (*)(void *))serve, csp);
3650 child_id = errno ? -1 : 0;
3651 pthread_attr_destroy(&attrs);
3655 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3656 #define SELECTED_ONE_OPTION
3657 child_id = _beginthread(
3658 (void (*)(void *))serve,
3663 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3664 #define SELECTED_ONE_OPTION
3665 child_id = _beginthread(
3666 (void(* _Optlink)(void*))serve,
3672 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3673 #define SELECTED_ONE_OPTION
3675 thread_id tid = spawn_thread
3676 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3678 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3680 child_id = (int) tid;
3689 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3690 #define SELECTED_ONE_OPTION
3691 csp->cfd = ReleaseSocket(csp->cfd, -1);
3694 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3695 NP_Output, Output(),
3696 NP_CloseOutput, FALSE,
3697 NP_Name, (ULONG)"privoxy child",
3701 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3702 NP_Output, Output(),
3703 NP_CloseOutput, FALSE,
3704 NP_Name, (ULONG)"privoxy child",
3705 NP_StackSize, 200*1024,
3711 ((struct Task *)child_id)->tc_UserData = csp;
3712 Signal((struct Task *)child_id, SIGF_SINGLE);
3717 #if !defined(SELECTED_ONE_OPTION)
3720 /* This block is only needed when using fork().
3721 * When using threads, the server thread was
3722 * created and run by the call to _beginthread().
3724 if (child_id == 0) /* child */
3727 #ifdef FEATURE_TOGGLE
3728 int inherited_toggle_state = global_toggle_state;
3729 #endif /* def FEATURE_TOGGLE */
3734 * If we've been toggled or we've blocked the request, tell Mom
3737 #ifdef FEATURE_TOGGLE
3738 if (inherited_toggle_state != global_toggle_state)
3740 rc |= RC_FLAG_TOGGLED;
3742 #endif /* def FEATURE_TOGGLE */
3744 #ifdef FEATURE_STATISTICS
3745 if (csp->flags & CSP_FLAG_REJECTED)
3747 rc |= RC_FLAG_BLOCKED;
3749 #endif /* ndef FEATURE_STATISTICS */
3753 else if (child_id > 0) /* parent */
3755 /* in a fork()'d environment, the parent's
3756 * copy of the client socket and the CSP
3760 #if !defined(_WIN32) && !defined(__CYGWIN__)
3762 wait( &child_status );
3765 * Evaluate child's return code: If the child has
3766 * - been toggled, toggle ourselves
3767 * - blocked its request, bump up the stats counter
3770 #ifdef FEATURE_TOGGLE
3771 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3773 global_toggle_state = !global_toggle_state;
3775 #endif /* def FEATURE_TOGGLE */
3777 #ifdef FEATURE_STATISTICS
3779 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3783 #endif /* def FEATURE_STATISTICS */
3785 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3786 close_socket(csp->cfd);
3787 csp->flags &= ~CSP_FLAG_ACTIVE;
3791 #undef SELECTED_ONE_OPTION
3792 /* end of cpp switch () */
3794 if (child_id < 0) /* failed */
3796 char buf[BUFFER_SIZE];
3798 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3800 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3802 write_socket(csp->cfd, buf, strlen(buf));
3803 close_socket(csp->cfd);
3804 csp->flags &= ~CSP_FLAG_ACTIVE;
3815 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3817 /* Clean up. Aim: free all memory (no leaks) */
3818 #ifdef FEATURE_GRACEFUL_TERMINATION
3820 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3822 unload_current_config_file();
3823 unload_current_actions_file();
3824 unload_current_re_filterfile();
3825 #ifdef FEATURE_TRUST
3826 unload_current_trust_file();
3829 if (config->multi_threaded)
3836 } while ((clients->next != NULL) && (--i > 0));
3840 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3851 #ifdef FEATURE_COOKIE_JAR
3852 if (NULL != config->jar)
3854 fclose(config->jar);
3858 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3859 /* Cleanup - remove taskbar icon etc. */
3864 #endif /* FEATURE_GRACEFUL_TERMINATION */