1 const char jcc_rcs[] = "$Id: jcc.c,v 1.143 2007/08/05 13:58:19 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 1.143 2007/08/05 13:58:19 fabiankeil
37 * Comment out request_contains_null_bytes() until it's used again.
39 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
40 * #1763173 from Stefan Huehner: s@const static@static const@
41 * and declare some more functions static.
43 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
44 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
45 * and explain why they were rejected in the first place.
46 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
47 * CONNECT requests. The request line was missing.
48 * - Add two more XXX reminders as we don't have enough already.
50 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
51 * As Hal noticed, checking dispatch_cgi() as the last cruncher
52 * looks like a bug if CGI requests are blocked unintentionally,
53 * so don't do it unless the user enabled the new config option
54 * "allow-cgi-request-crunching".
56 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
57 * - Allow to rewrite the request destination behind the client's back.
58 * - Turn the weird-looking unconditional for loop that
59 * reads the client request into a conditional while loop.
60 * Move the stuff that only runs once out of the loop.
61 * - Move parts of chat(), server_content_type() and the
62 * necessary stuff to fix BR#1750917 into get_filter_function().
64 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
65 * Temporary workaround for BR#1730105.
67 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
68 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
69 * deadlocks and crashes on OpenBSD and possibly other OS with neither
70 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
71 * Thanks to Ralf Horstmann for report and solution.
73 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
74 * Add forward-override{} to change the forwarding settings through
75 * action sections. This is mainly interesting to forward different
76 * clients differently (for example based on User-Agent or request
79 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
80 * - Let usage() mention the --chroot parameter.
81 * - Use read_socket() consistently and always leave
82 * the last buffer byte alone, even in cases where
83 * null termination (currently) doesn't matter.
85 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
86 * - Fix config file loading on Unix if no config file is specified.
87 * Since r1.97 Privoxy would always interpret the last argument as
88 * config file, even if it's a valid command line option.
89 * - Abort in case of unrecognized command line options. Closes #1719696.
90 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
91 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
93 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
94 * - Don't rerun crunchers that only depend on the request URL.
95 * - Don't count redirects and CGI requests as "blocked requests".
97 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
98 * Support crunching based on tags created by server-header taggers.
100 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
101 * Make HTTP snippets static (again). Add a Content-Type for those
102 * with content so the browser doesn't guess it based on the URL.
104 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
105 * Move crunching and request line rebuilding out of chat().
107 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
108 * Introduce tags as alternative way to specify which
109 * actions apply to a request. At the moment tags can be
110 * created based on client and server headers.
112 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
113 * Don't CLF-log CONNECT requests twice.
115 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
116 * Log the source address for ACL-related connection drops.
118 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
119 * New config option: enforce-blocks.
121 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
122 * - Move null byte check into separate function.
123 * - Don't confuse the client with error pages
124 * if a CONNECT request was already confirmed.
126 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
127 * Speed up NULL byte escaping and only log the complete
128 * NULL byte requests with header debugging enabled.
130 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
131 * Answer requests that contain NULL bytes with
132 * a custom response instead of waiting for more
133 * data until the client eventually hangs up.
135 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
136 * - Move delivery and logging of crunched responses
137 * from chat() into send_crunch_response().
138 * - Display the reason for generating http_responses.
139 * - Log the content length for LOG_LEVEL_CLF correctly
140 * (still incorrect for some fixed responses).
141 * - Reword an incorrect comment about
142 * treat-forbidden-connects-like-blocks violating
144 * - Add some log messages.
146 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
147 * Move mutex initialization into separate
148 * function and exit in case of errors.
150 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
151 * - Start to reduce chat()'s line count and move
152 * parts of it into separate functions.
153 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
155 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
156 * - Add Proxy-Agent header to HTTP snippets that are
157 * supposed to reach HTTP clients only.
158 * - Made a few CONNECT log messages more descriptive.
159 * - Catch completely empty server responses (as seen
160 * with Tor's fake ".noconnect" top level domain).
161 * - Use shiny new "forwarding-failed" template for socks errors.
163 * Revision 1.118 2007/01/07 07:43:43 joergs
164 * AmigaOS4 support added.
166 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
167 * Added config option accept-intercepted-requests
168 * and disabled it by default.
170 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
171 * Reverted parts of my last commit
172 * to keep error handling working.
174 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
175 * Fixed gcc43 conversion warnings.
177 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
178 * Fix -pedantic ISO C warning about converting
179 * from function pointer to object pointer.
181 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
182 * Silence compiler warning I introduced with my last commit.
184 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
185 * Mutex protect rand() if POSIX threading
186 * is used, warn the user if that's not possible
187 * and stop using it on _WIN32 where it could
190 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
191 * Don't prevent core dumps by catching SIGABRT.
192 * It's rude and makes debugging unreasonable painful.
194 * Revision 1.110 2006/12/13 14:52:53 etresoft
195 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
197 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
198 * Privoxy is now able to run as intercepting
199 * proxy in combination with any packet filter
200 * that does the port redirection. The destination
201 * is extracted from the "Host:" header which
202 * should be available for nearly all requests.
204 * Moved HTTP snipplets into jcc.c.
205 * Added error message for gopher proxy requests.
207 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
208 * Only unlink the pidfile if it's actually used.
210 * Change order of interception checks to make
211 * it possible to block or redirect requests for
214 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
215 * Make pthread mutex locking more generic. Instead of
216 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
217 * and use mutex locking unless there is an _r function
218 * available. Better safe than sorry.
220 * Fixes "./configure --disable-pthread" and should result
221 * in less threading-related problems on pthread-using platforms,
222 * but it still doesn't fix BR#1122404.
224 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
225 * Move pthread.h inclusion from jcc.c to jcc.h.
226 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
228 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
229 * Don't exit after receiving the second SIGHUP on Solaris.
231 * Fixes BR 1052235, but the same problem may exist on other
232 * systems. Once 3.0.6 is out we should use sigset()
233 * where available and see if it breaks anything.
235 * Revision 1.104 2006/09/23 13:26:38 roro
236 * Replace TABs by spaces in source code.
238 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
239 * Fix +redirect{}. Didn't work with -fast-redirects.
241 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
242 * Respond with 400 and a short text message
243 * if the client tries to use Privoxy as FTP proxy.
245 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
246 * Make number of retries in case of forwarded-connect problems
247 * a config file option (forwarded-connect-retries) and use 0 as
250 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
251 * Set random(3) seed.
253 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
254 * Follow the OpenBSD port's lead and protect the resolve
255 * functions on OpenBSD as well.
257 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
258 * --user fix. Only use the user as group if no group is specified.
259 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
261 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
262 * Windows service (re-)integration
266 * --install[:service_name]
267 * --uninstall[:service_name]
270 * They work as follows:
271 * --install will create a service for you and then terminate.
272 * By default the service name will be "privoxy" (without the quotes).
273 * However you can run multiple services if you wish, just by adding
274 * a colon and then a name (no spaces).
276 * --uninstall follows the exact same rules a --install.
278 * --service is used when the program is executed by the service
279 * control manager, and in normal circumstances would never be
280 * used as a command line argument.
282 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
283 * Windows service integration
285 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
286 * Incorporate Fabian Keil's patch work:
287 http://www.fabiankeil.de/sourcecode/privoxy/
289 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
290 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
291 * with what was really the latest development (the v_3_0_branch branch)
293 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
294 * Thanks to Jindrich Makovicka for a race condition fix for the log
295 * file. The race condition remains for non-pthread implementations.
296 * Reference patch #1175720.
298 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
299 * Add the ability to check jpeg images for invalid
300 * lengths of comment blocks. Defensive strategy
301 * against the exploit:
302 * Microsoft Security Bulletin MS04-028
303 * Buffer Overrun in JPEG Processing (GDI+) Could
304 * Allow Code Execution (833987)
305 * Enabled with +inspect-jpegs in actions files.
307 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
308 * - Fixed usage info for non-unix platforms
309 * - Fixed small cmdline parsing bug
311 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
312 * Diagnostics: Now preserve the returncode of pthread_create
313 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
315 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
316 * No longer ignore SIGCHLD. Fixes bug #769381
318 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
319 * Close jarfile on graceful exit, remove stray line
321 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
322 * Cosmetics: Killed a warning, a typo and an allocation left at exit
324 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
325 * No longer rely on non-POSIX.1 extensions of getcwd().
328 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
329 * Replaced setenv() by posix-compliant putenv()
330 * Thanks to Neil McCalden (nmcc AT users.sf.net).
332 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
333 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
335 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
336 * Clean-up and extension of improvements for forked mode:
337 * - Child's return code now consists of flags RC_FLAG_*
338 * - Reporting toggle to parent now properly #ifdef'ed
339 * - Children now report blocking to parent. This enables
340 * statistics in forked mode
342 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
343 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
344 * Children now report having been toggled through _exit(17), parents
345 * watch for that code and toggle themselves if found.
347 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
348 * Wrapping all *_r functions (the non-_r versions of them) with
349 * mutex semaphores for OSX. Hopefully this will take care of all
350 * of those pesky crash reports.
352 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
353 * Fixed two mostly harmless mem leaks
355 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
356 * Fix: Head of global clients list now initialized to NULL
358 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
359 * Added basic support for OPTIONS and TRACE HTTP methods:
360 * - New interceptor direct_response() added in chat().
361 * - sed() moved to earlier in the process, so that the
362 * Host: header is evaluated before actions and forwarding
365 * Revision 1.92 2002/05/08 16:00:46 oes
366 * Chat's buffer handling:
367 * - Fixed bug with unchecked out-of-mem conditions
368 * while reading client request & server headers
369 * - No longer predict if the buffer limit will be exceeded
370 * in the next read -- check add_to_iob's new
371 * return code. If buffer couldn't be extended
372 * (policy or out-of-mem) while
373 * - reading from client: abort
374 * - reading server headers: send error page
375 * - buffering server body for filter: flush,
376 * and if that fails: send error page
378 * Revision 1.91 2002/04/08 20:35:58 swa
381 * Revision 1.90 2002/04/02 14:57:28 oes
382 * Made sending wafers independent of FEATURE_COOKIE_JAR
384 * Revision 1.89 2002/03/31 17:18:59 jongfoster
385 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
387 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
388 * More compiler warning message maintenance
390 * Revision 1.87 2002/03/26 22:29:54 swa
391 * we have a new homepage!
393 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
394 * Workaround for closing the jarfile before load_config() comes around again
396 * Revision 1.85 2002/03/24 15:23:33 jongfoster
399 * Revision 1.84 2002/03/24 13:25:43 swa
400 * name change related issues
402 * Revision 1.83 2002/03/16 23:54:06 jongfoster
403 * Adding graceful termination feature, to help look for memory leaks.
404 * If you enable this (which, by design, has to be done by hand
405 * editing config.h) and then go to http://i.j.b/die, then the program
406 * will exit cleanly after the *next* request. It should free all the
407 * memory that was used.
409 * Revision 1.82 2002/03/13 00:27:05 jongfoster
412 * Revision 1.81 2002/03/12 01:42:50 oes
413 * Introduced modular filters
415 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
416 * OS/2 port maintenance:
417 * - Fixed EMX build - it had decayed a little
418 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
419 * substituted a memset for now.
421 * Revision 1.79 2002/03/09 20:03:52 jongfoster
422 * - Making various functions return int rather than size_t.
423 * (Undoing a recent change). Since size_t is unsigned on
424 * Windows, functions like read_socket that return -1 on
425 * error cannot return a size_t.
427 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
428 * crashes, and also frequently caused JB to jump to 100%
429 * CPU and stay there. (Because it thought it had just
430 * read ((unsigned)-1) == 4Gb of data...)
432 * - The signature of write_socket has changed, it now simply
433 * returns success=0/failure=nonzero.
435 * - Trying to get rid of a few warnings --with-debug on
436 * Windows, I've introduced a new type "jb_socket". This is
437 * used for the socket file descriptors. On Windows, this
438 * is SOCKET (a typedef for unsigned). Everywhere else, it's
439 * an int. The error value can't be -1 any more, so it's
440 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
441 * Windows it maps to the #define INVALID_SOCKET.)
443 * - The signature of bind_port has changed.
445 * Revision 1.78 2002/03/08 21:35:04 oes
446 * Added optional group supplement to --user option. Will now use default group of user if no group given
448 * Revision 1.77 2002/03/07 03:52:06 oes
449 * - Fixed compiler warnings etc
450 * - Improved handling of failed DNS lookups
452 * Revision 1.76 2002/03/06 22:54:35 jongfoster
453 * Automated function-comment nitpicking.
455 * Revision 1.75 2002/03/06 10:02:19 oes
456 * Fixed stupid bug when --user was not given
458 * Revision 1.74 2002/03/06 00:49:31 jongfoster
459 * Fixing warning on Windows
460 * Making #ifdefs that refer to the same variable consistently
461 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
463 * Revision 1.73 2002/03/05 23:57:30 hal9
464 * Stray character 's' on line 1618 was breaking build.
466 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
467 * - Re-enable OS/2 building after new parms were added
468 * - Fix false out of memory report when resolving CGI templates when no IP
469 * address is available of failed attempt (a la no such domain)
471 * Revision 1.71 2002/03/05 18:13:56 oes
472 * Added --user option
474 * Revision 1.70 2002/03/05 04:52:42 oes
475 * Deleted non-errlog debugging code
477 * Revision 1.69 2002/03/04 23:50:00 jongfoster
478 * Splitting off bind_port() call into bind_port_helper(), with
481 * Revision 1.68 2002/03/04 20:17:32 oes
484 * Revision 1.67 2002/03/04 18:18:57 oes
485 * - Removed _DEBUG mode
486 * - Cleand up cmdline parsing
487 * - Introduced --no-daemon, --pidfile options
488 * - Cleaned up signal handling:
489 * - Terminate cleanly on INT, TERM and ABRT
490 * - Schedule logfile for re-opening on HUP
491 * - Ignore CHLD and PIPE
492 * - Leave the rest with their default handlers
493 * - Uniform handler registration
494 * - Added usage() function
495 * - Played styleguide police
497 * Revision 1.66 2002/03/03 15:06:55 oes
498 * Re-enabled automatic config reloading
500 * Revision 1.65 2002/03/03 14:49:11 oes
501 * Fixed CLF logging: Now uses client's original HTTP request
503 * Revision 1.64 2002/03/03 09:18:03 joergs
504 * Made jumbjuster work on AmigaOS again.
506 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
507 * Clean up a little CRLF unpleasantness that suddenly appeared
509 * Revision 1.62 2002/02/20 23:17:23 jongfoster
510 * Detecting some out-of memory conditions and exiting with a log message.
512 * Revision 1.61 2002/01/17 21:01:52 jongfoster
513 * Moving all our URL and URL pattern parsing code to urlmatch.c.
515 * Revision 1.60 2001/12/30 14:07:32 steudten
516 * - Add signal handling (unix)
517 * - Add SIGHUP handler (unix)
518 * - Add creation of pidfile (unix)
519 * - Add action 'top' in rc file (RH)
520 * - Add entry 'SIGNALS' to manpage
521 * - Add exit message to logfile (unix)
523 * Revision 1.59 2001/12/13 14:07:18 oes
524 * Fixed Bug: 503 error page now sent OK
526 * Revision 1.58 2001/11/30 23:37:24 jongfoster
527 * Renaming the Win32 config file to config.txt - this is almost the
528 * same as the corresponding UNIX name "config"
530 * Revision 1.57 2001/11/16 00:47:43 jongfoster
531 * Changing the tty-disconnection code to use setsid().
533 * Revision 1.56 2001/11/13 20:20:54 jongfoster
534 * Tabs->spaces, fixing a bug with missing {} around an if()
536 * Revision 1.55 2001/11/13 20:14:53 jongfoster
537 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
539 * Revision 1.54 2001/11/07 00:03:14 steudten
540 * Give reliable return value if an error
541 * occurs not just 0 with new daemon mode.
543 * Revision 1.53 2001/11/05 21:41:43 steudten
544 * Add changes to be a real daemon just for unix os.
545 * (change cwd to /, detach from controlling tty, set
546 * process group and session leader to the own process.
548 * Add some fatal-error log message for failed malloc().
549 * Add '-d' if compiled with 'configure --with-debug' to
550 * enable debug output.
552 * Revision 1.52 2001/10/26 20:11:20 jongfoster
553 * Fixing type mismatch
555 * Revision 1.51 2001/10/26 17:38:28 oes
558 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
559 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
560 * threads to call select() simultaneously. So, it's time to do a real, live,
561 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
562 * (native). Both versions will work, but using __OS2__ offers multi-threading.
564 * Revision 1.49 2001/10/23 21:41:35 jongfoster
565 * Added call to initialize the (statically-allocated of course)
566 * "out of memory" CGI response.
568 * Revision 1.48 2001/10/10 19:56:46 jongfoster
569 * Moving some code that wasn't cookie-related out of an #ifdef
572 * Revision 1.47 2001/10/10 16:44:36 oes
573 * Added CONNECT destination port limitation check
575 * Revision 1.46 2001/10/08 15:17:41 oes
576 * Re-enabled SSL forwarding
578 * Revision 1.45 2001/10/07 15:42:11 oes
579 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
581 * Moved downgrading of the HTTP version from parse_http_request to
582 * chat(), since we can't decide if it is necessary before we have
583 * determined the actions for the URL. The HTTP command is now
584 * *always* re-built so the repairs need no longer be special-cased.
586 * filter_popups now gets a csp pointer so it can raise the new
587 * CSP_FLAG_MODIFIED flag.
591 * Added configurable size limit for the IOB. If the IOB grows so
592 * large that the next read would exceed the limit, the header
593 * is generated, and the header & unfiltered buffer are flushed
594 * to the client. Chat then continues in non-buffering,
595 * non-filtering body mode.
597 * Revision 1.44 2001/10/02 18:13:57 oes
600 * Revision 1.43 2001/10/02 15:32:13 oes
601 * Moved generation of hdr
603 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
604 * Cleaning up 2 compiler warnings on OS/2.
606 * Revision 1.41 2001/09/16 17:05:14 jongfoster
607 * Removing unused #include showarg.h
609 * Revision 1.40 2001/09/16 15:41:45 jongfoster
610 * Fixing signed/unsigned comparison warning.
612 * Revision 1.39 2001/09/16 13:21:27 jongfoster
613 * Changes to use new list functions.
615 * Revision 1.38 2001/09/16 13:01:46 jongfoster
616 * Removing redundant function call that zeroed zalloc()'d memory.
618 * Revision 1.37 2001/09/10 11:12:24 oes
619 * Deleted unused variable
621 * Revision 1.36 2001/09/10 10:56:15 oes
622 * Silenced compiler warnings
624 * Revision 1.35 2001/07/31 14:44:22 oes
625 * Deleted unused size parameter from filter_popups()
627 * Revision 1.34 2001/07/30 22:08:36 jongfoster
628 * Tidying up #defines:
629 * - All feature #defines are now of the form FEATURE_xxx
630 * - Permanently turned off WIN_GUI_EDIT
631 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
633 * Revision 1.33 2001/07/29 19:32:00 jongfoster
634 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
636 * Revision 1.32 2001/07/29 18:47:05 jongfoster
637 * Adding missing #include "loadcfg.h"
639 * Revision 1.31 2001/07/29 12:17:48 oes
640 * Applied pthread fix by Paul Lieverse
642 * Revision 1.30 2001/07/25 22:57:13 jongfoster
643 * __BEOS__ no longer overrides FEATURE_PTHREAD.
644 * This is because FEATURE_PTHREAD will soon be widely used, so I
645 * want to keep it simple.
647 * Revision 1.29 2001/07/24 12:47:06 oes
648 * Applied BeOS support update by Eugenia
650 * Revision 1.28 2001/07/23 13:26:12 oes
651 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
653 * Revision 1.27 2001/07/19 19:09:47 haroon
654 * - Added code to take care of the situation where while processing the first
655 * server response (which includes the server header), after finding the end
656 * of the headers we were not looking past the end of the headers for
657 * content modification. I enabled it for filter_popups.
658 * Someone else should look to see if other similar operations should be
659 * done to the discarded portion of the buffer.
661 * Note 2001/07/20: No, the other content modification mechanisms will process
662 * the whole iob later anyway. --oes
664 * Revision 1.26 2001/07/18 12:31:36 oes
667 * Revision 1.25 2001/07/15 19:43:49 jongfoster
668 * Supports POSIX threads.
669 * Also removed some unused #includes.
671 * Revision 1.24 2001/07/13 14:00:40 oes
672 * - Generic content modification scheme:
673 * Each feature has its own applicability flag that is set
674 * from csp->action->flags.
675 * Replaced the "filtering" int flag , by a function pointer
676 * "content_filter" to the function that will do the content
677 * modification. If it is != NULL, the document will be buffered
678 * and processed through *content_filter, which must set
679 * csp->content_length and return a modified copy of the body
680 * or return NULL (on failiure).
681 * - Changed csp->is_text to the more generic bitmap csp->content_type
682 * which can currently take the valued CT_TEXT or CT_GIF
684 * - Removed all #ifdef PCRS
686 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
687 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
688 * the `block_acl' call. This prevents a compilation error when the user
689 * does not wish to use the "ACL" feature.
691 * Revision 1.22 2001/06/29 21:45:41 oes
692 * Indentation, CRLF->LF, Tab-> Space
694 * Revision 1.21 2001/06/29 13:29:36 oes
695 * - Cleaned up, improved comments
696 * - Unified all possible interceptors (CGI,
697 * block, trust, fast_redirect) in one
698 * place, with one (CGI) answer generation
699 * mechansim. Much clearer now.
700 * - Removed the GIF image generation, which
701 * is now done in filters.c:block_url()
702 * - Made error conditions like domain lookup
703 * failiure or (various) problems while talking
704 * to the server use cgi.c:error_response()
705 * instead of generating HTML/HTTP in chat() (yuck!)
706 * - Removed logentry from cancelled commit
708 * Revision 1.20 2001/06/09 10:55:28 jongfoster
709 * Changing BUFSIZ ==> BUFFER_SIZE
711 * Revision 1.19 2001/06/07 23:12:52 jongfoster
712 * Replacing function pointer in struct gateway with a directly
713 * called function forwarded_connect().
714 * Replacing struct gateway with struct forward_spec
716 * Revision 1.18 2001/06/03 19:12:16 oes
717 * introduced new cgi handling
719 * Revision 1.17 2001/06/01 20:07:23 jongfoster
720 * Now uses action +image-blocker{} rather than config->tinygif
722 * Revision 1.16 2001/06/01 18:49:17 jongfoster
723 * Replaced "list_share" with "list" - the tiny memory gain was not
724 * worth the extra complexity.
726 * Revision 1.15 2001/05/31 21:24:47 jongfoster
727 * Changed "permission" to "action" throughout.
728 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
729 * Moved vanilla wafer check into chat(), since we must now
730 * decide whether or not to add it based on the URL.
732 * Revision 1.14 2001/05/29 20:14:01 joergs
733 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
736 * Revision 1.13 2001/05/29 09:50:24 jongfoster
737 * Unified blocklist/imagelist/permissionslist.
738 * File format is still under discussion, but the internal changes
741 * Also modified interceptor behaviour:
742 * - We now intercept all URLs beginning with one of the following
743 * prefixes (and *only* these prefixes):
745 * * http://ijbswa.sf.net/config/
746 * * http://ijbswa.sourceforge.net/config/
747 * - New interceptors "home page" - go to http://i.j.b/ to see it.
748 * - Internal changes so that intercepted and fast redirect pages
749 * are not replaced with an image.
750 * - Interceptors now have the option to send a binary page direct
751 * to the client. (i.e. ijb-send-banner uses this)
752 * - Implemented show-url-info interceptor. (Which is why I needed
753 * the above interceptors changes - a typical URL is
754 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
755 * The previous mechanism would not have intercepted that, and
756 * if it had been intercepted then it then it would have replaced
759 * Revision 1.12 2001/05/27 22:17:04 oes
761 * - re_process_buffer no longer writes the modified buffer
762 * to the client, which was very ugly. It now returns the
763 * buffer, which it is then written by chat.
765 * - content_length now adjusts the Content-Length: header
766 * for modified documents rather than crunch()ing it.
767 * (Length info in csp->content_length, which is 0 for
768 * unmodified documents)
770 * - For this to work, sed() is called twice when filtering.
772 * Revision 1.11 2001/05/26 17:27:53 jongfoster
773 * Added support for CLF and fixed LOG_LEVEL_LOG.
774 * Also did CRLF->LF fix of my previous patch.
776 * Revision 1.10 2001/05/26 15:26:15 jongfoster
777 * ACL feature now provides more security by immediately dropping
778 * connections from untrusted hosts.
780 * Revision 1.9 2001/05/26 00:28:36 jongfoster
781 * Automatic reloading of config file.
782 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
783 * Most of the global variables have been moved to a new
784 * struct configuration_spec, accessed through csp->config->globalname
785 * Most of the globals remaining are used by the Win32 GUI.
787 * Revision 1.8 2001/05/25 22:43:18 jongfoster
788 * Fixing minor memory leak and buffer overflow.
790 * Revision 1.7 2001/05/25 22:34:30 jongfoster
793 * Revision 1.6 2001/05/23 00:13:58 joergs
794 * AmigaOS support fixed.
796 * Revision 1.5 2001/05/22 18:46:04 oes
798 * - Enabled filtering banners by size rather than URL
799 * by adding patterns that replace all standard banner
800 * sizes with the "Junkbuster" gif to the re_filterfile
802 * - Enabled filtering WebBugs by providing a pattern
803 * which kills all 1x1 images
805 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
806 * which is selected by the (nonstandard and therefore
807 * capital) letter 'U' in the option string.
808 * It causes the quantifiers to be ungreedy by default.
809 * Appending a ? turns back to greedy (!).
811 * - Added a new interceptor ijb-send-banner, which
812 * sends back the "Junkbuster" gif. Without imagelist or
813 * MSIE detection support, or if tinygif = 1, or the
814 * URL isn't recognized as an imageurl, a lame HTML
815 * explanation is sent instead.
817 * - Added new feature, which permits blocking remote
818 * script redirects and firing back a local redirect
820 * The feature is conditionally compiled, i.e. it
821 * can be disabled with --disable-fast-redirects,
822 * plus it must be activated by a "fast-redirects"
823 * line in the config file, has its own log level
824 * and of course wants to be displayed by show-proxy-args
825 * Note: Boy, all the #ifdefs in 1001 locations and
826 * all the fumbling with configure.in and acconfig.h
827 * were *way* more work than the feature itself :-(
829 * - Because a generic redirect template was needed for
830 * this, tinygif = 3 now uses the same.
832 * - Moved GIFs, and other static HTTP response templates
837 * - Removed some >400 CRs again (Jon, you really worked
840 * Revision 1.4 2001/05/21 19:34:01 jongfoster
841 * Made failure to bind() a fatal error.
843 * Revision 1.3 2001/05/20 01:21:20 jongfoster
844 * Version 2.9.4 checkin.
845 * - Merged popupfile and cookiefile, and added control over PCRS
846 * filtering, in new "permissionsfile".
847 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
848 * file error you now get a message box (in the Win32 GUI) rather
849 * than the program exiting with no explanation.
850 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
852 * - Removed tabs from "config"
853 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
854 * - Bumped up version number.
856 * Revision 1.2 2001/05/17 22:34:44 oes
857 * - Added hint on GIF char array generation to jcc.c
858 * - Cleaned CRLF's from the sources and related files
859 * - Repaired logging for REF and FRC
861 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
862 * Initial import of version 2.9.3 source tree
865 *********************************************************************/
871 #include <sys/types.h>
880 # ifndef FEATURE_PTHREAD
884 # include <windows.h>
885 # include <process.h>
886 # endif /* ndef FEATURE_PTHREAD */
889 # ifndef _WIN_CONSOLE
891 # endif /* ndef _WIN_CONSOLE */
892 # include "w32svrapi.h"
894 #else /* ifndef _WIN32 */
896 # if !defined (__OS2__)
898 # include <sys/wait.h>
899 # endif /* ndef __OS2__ */
900 # include <sys/time.h>
901 # include <sys/stat.h>
902 # include <sys/ioctl.h>
905 #include <sys/termios.h>
916 # include <socket.h> /* BeOS has select() for sockets only. */
917 # include <OS.h> /* declarations for threads and stuff. */
920 # if defined(__EMX__) || defined(__OS2__)
921 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
926 #define bzero(B,N) memset(B,0x00,n)
941 #include "killpopup.h"
942 #include "miscutil.h"
944 #include "jbsockets.h"
949 #include "urlmatch.h"
951 const char jcc_h_rcs[] = JCC_H_VERSION;
952 const char project_h_rcs[] = PROJECT_H_VERSION;
955 struct client_state clients[1];
956 struct file_list files[1];
958 #ifdef FEATURE_STATISTICS
959 int urls_read = 0; /* total nr of urls read inc rejected */
960 int urls_rejected = 0; /* total nr of urls rejected */
961 #endif /* def FEATURE_STATISTICS */
963 #ifdef FEATURE_GRACEFUL_TERMINATION
967 static void sig_handler(int the_signal);
968 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
969 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
970 static jb_err get_server_headers(struct client_state *csp);
971 static const char *crunch_reason(const struct http_response *rsp);
972 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
974 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
976 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
977 static jb_err change_request_destination(struct client_state *csp);
978 static void chat(struct client_state *csp);
979 static void serve(struct client_state *csp);
980 static void usage(const char *myname);
981 static void initialize_mutexes(void);
982 static jb_socket bind_port_helper(struct configuration_spec *config);
983 static void listen_loop(void);
986 void serve(struct client_state *csp);
987 #else /* ifndef AMIGA */
988 static void serve(struct client_state *csp);
989 #endif /* def AMIGA */
992 static int32 server_thread(void *data);
993 #endif /* def __BEOS__ */
996 #define sleep(N) Sleep(((N) * 1000))
1000 #define sleep(N) DosSleep(((N) * 100))
1003 #ifdef FEATURE_PTHREAD
1004 pthread_mutex_t log_mutex;
1005 pthread_mutex_t log_init_mutex;
1007 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1008 pthread_mutex_t resolver_mutex;
1009 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1011 #ifndef HAVE_GMTIME_R
1012 pthread_mutex_t gmtime_mutex;
1013 #endif /* ndef HAVE_GMTIME_R */
1015 #ifndef HAVE_LOCALTIME_R
1016 pthread_mutex_t localtime_mutex;
1017 #endif /* ndef HAVE_GMTIME_R */
1020 pthread_mutex_t rand_mutex;
1021 #endif /* ndef HAVE_RANDOM */
1023 #endif /* FEATURE_PTHREAD */
1025 #if defined(unix) || defined(__EMX__)
1026 const char *basedir = NULL;
1027 const char *pidfile = NULL;
1028 int received_hup_signal = 0;
1029 #endif /* defined unix */
1031 /* The vanilla wafer. */
1032 static const char VANILLA_WAFER[] =
1033 "NOTICE=TO_WHOM_IT_MAY_CONCERN_"
1034 "Do_not_send_me_any_copyrighted_information_other_than_the_"
1035 "document_that_I_am_requesting_or_any_of_its_necessary_components._"
1036 "In_particular_do_not_send_me_any_cookies_that_"
1037 "are_subject_to_a_claim_of_copyright_by_anybody._"
1038 "Take_notice_that_I_refuse_to_be_bound_by_any_license_condition_"
1039 "(copyright_or_otherwise)_applying_to_any_cookie._";
1041 /* HTTP snipplets. */
1042 static const char CSUCCEED[] =
1043 "HTTP/1.0 200 Connection established\n"
1044 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1046 static const char CHEADER[] =
1047 "HTTP/1.0 400 Invalid header received from browser\r\n"
1048 "Proxy-Agent: Privoxy " VERSION "\r\n"
1049 "Content-Type: text/plain\r\n"
1050 "Connection: close\r\n\r\n"
1051 "Invalid header received from browser.\r\n";
1053 static const char CFORBIDDEN[] =
1054 "HTTP/1.0 403 Connection not allowable\r\n"
1055 "Proxy-Agent: Privoxy " VERSION "\r\n"
1056 "X-Hint: If you read this message interactively, then you know why this happens ,-)\r\n"
1057 "Connection: close\r\n\r\n";
1059 static const char FTP_RESPONSE[] =
1060 "HTTP/1.0 400 Invalid request received from browser\r\n"
1061 "Content-Type: text/plain\r\n"
1062 "Connection: close\r\n\r\n"
1063 "Invalid request. Privoxy doesn't support FTP.\r\n";
1065 static const char GOPHER_RESPONSE[] =
1066 "HTTP/1.0 400 Invalid request received from browser\r\n"
1067 "Content-Type: text/plain\r\n"
1068 "Connection: close\r\n\r\n"
1069 "Invalid request. Privoxy doesn't support gopher.\r\n";
1071 /* XXX: should be a template */
1072 static const char MISSING_DESTINATION_RESPONSE[] =
1073 "HTTP/1.0 400 Bad request received from browser\r\n"
1074 "Proxy-Agent: Privoxy " VERSION "\r\n"
1075 "Content-Type: text/plain\r\n"
1076 "Connection: close\r\n\r\n"
1077 "Bad request. Privoxy was unable to extract the destination.\r\n";
1079 /* XXX: should be a template */
1080 static const char NO_SERVER_DATA_RESPONSE[] =
1081 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1082 "Proxy-Agent: Privoxy " VERSION "\r\n"
1083 "Content-Type: text/plain\r\n"
1084 "Connection: close\r\n\r\n"
1085 "Empty server or forwarder response.\r\n"
1086 "The connection was closed without sending any data.\r\n";
1088 /* XXX: should be a template */
1089 static const char NULL_BYTE_RESPONSE[] =
1090 "HTTP/1.0 400 Bad request received from browser\r\n"
1091 "Proxy-Agent: Privoxy " VERSION "\r\n"
1092 "Content-Type: text/plain\r\n"
1093 "Connection: close\r\n\r\n"
1094 "Bad request. Null byte(s) before end of request.\r\n";
1096 /* XXX: should be a template */
1097 static const char MESSED_UP_REQUEST_RESPONSE[] =
1098 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1099 "Proxy-Agent: Privoxy " VERSION "\r\n"
1100 "Content-Type: text/plain\r\n"
1101 "Connection: close\r\n\r\n"
1102 "Bad request. Messed up with header filters.\r\n";
1104 /* A function to crunch a response */
1105 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1107 typedef char *(*filter_function_ptr)();
1109 /* Crunch function flags */
1110 #define CF_NO_FLAGS 0
1111 /* Cruncher applies to forced requests as well */
1112 #define CF_IGNORE_FORCE 1
1113 /* Crunched requests are counted for the block statistics */
1114 #define CF_COUNT_AS_REJECT 2
1116 /* A crunch function and its flags */
1119 const crunch_func_ptr cruncher;
1123 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1124 static filter_function_ptr get_filter_function(struct client_state *csp);
1126 /* Complete list of cruncher functions */
1127 static const struct cruncher crunchers_all[] = {
1128 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1129 { block_url, CF_COUNT_AS_REJECT },
1130 #ifdef FEATURE_TRUST
1131 { trust_url, CF_COUNT_AS_REJECT },
1132 #endif /* def FEATURE_TRUST */
1133 { redirect_url, CF_NO_FLAGS },
1134 { dispatch_cgi, CF_IGNORE_FORCE},
1138 /* Light version, used after tags are applied */
1139 static const struct cruncher crunchers_light[] = {
1140 { block_url, CF_COUNT_AS_REJECT },
1141 { redirect_url, CF_NO_FLAGS },
1146 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1147 /*********************************************************************
1149 * Function : sig_handler
1151 * Description : Signal handler for different signals.
1152 * Exit gracefully on TERM and INT
1153 * or set a flag that will cause the errlog
1154 * to be reopened by the main thread on HUP.
1157 * 1 : the_signal = the signal cause this function to call
1161 *********************************************************************/
1162 static void sig_handler(int the_signal)
1168 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1179 received_hup_signal = 1;
1184 * We shouldn't be here, unless we catch signals
1185 * in main() that we can't handle here!
1187 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1195 /*********************************************************************
1197 * Function : client_protocol_is_unsupported
1199 * Description : Checks if the client used a known unsupported
1200 * protocol and deals with it by sending an error
1204 * 1 : csp = Current client state (buffers, headers, etc...)
1205 * 2 : req = the first request line send by the client
1207 * Returns : TRUE if an error response has been generated, or
1208 * FALSE if the request doesn't look invalid.
1210 *********************************************************************/
1211 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1213 char buf[BUFFER_SIZE];
1216 * If it's a FTP or gopher request, we don't support it.
1218 * These checks are better than nothing, but they might
1219 * not work in all configurations and some clients might
1220 * have problems digesting the answer.
1222 * They should, however, never cause more problems than
1223 * Privoxy's old behaviour (returning the misleading HTML
1226 * "Could not resolve http://(ftp|gopher)://example.org").
1228 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1230 if (!strncmpic(req, "GET ftp://", 10))
1232 strlcpy(buf, FTP_RESPONSE, sizeof(buf));
1233 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1234 csp->ip_addr_str, req);
1238 strlcpy(buf, GOPHER_RESPONSE, sizeof(buf));
1239 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1240 csp->ip_addr_str, req);
1242 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1244 write_socket(csp->cfd, buf, strlen(buf));
1253 /*********************************************************************
1255 * Function : get_request_destination_elsewhere
1257 * Description : If the client's request was redirected into
1258 * Privoxy without the client's knowledge,
1259 * the request line lacks the destination host.
1261 * This function tries to get it elsewhere,
1262 * provided accept-intercepted-requests is enabled.
1264 * "Elsewhere" currently only means "Host: header",
1265 * but in the future we may ask the redirecting
1266 * packet filter to look the destination up.
1268 * If the destination stays unknown, an error
1269 * response is send to the client and headers
1270 * are freed so that chat() can return directly.
1273 * 1 : csp = Current client state (buffers, headers, etc...)
1274 * 2 : headers = a header list
1276 * Returns : JB_ERR_OK if the destination is now known, or
1277 * JB_ERR_PARSE if it isn't.
1279 *********************************************************************/
1280 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1284 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1286 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1287 " Privoxy isn't configured to accept intercepted requests.",
1288 csp->ip_addr_str, csp->http->cmd);
1289 /* XXX: Use correct size */
1290 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1291 csp->ip_addr_str, csp->http->cmd);
1293 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1294 destroy_list(headers);
1296 return JB_ERR_PARSE;
1298 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1300 /* Split the domain we just got for pattern matching */
1301 init_domain_components(csp->http);
1307 /* We can't work without destination. Go spread the news.*/
1309 req = list_to_text(headers);
1311 /* XXX: Use correct size */
1312 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1313 csp->ip_addr_str, csp->http->cmd);
1314 log_error(LOG_LEVEL_ERROR,
1315 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1316 csp->ip_addr_str, csp->http->cmd, req);
1319 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1320 destroy_list(headers);
1322 return JB_ERR_PARSE;
1325 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1326 * to get the destination IP address, use it as host directly
1327 * or do a reverse DNS lookup first.
1332 /*********************************************************************
1334 * Function : get_server_headers
1336 * Description : Parses server headers in iob and fills them
1337 * into csp->headers so that they can later be
1341 * 1 : csp = Current client state (buffers, headers, etc...)
1343 * Returns : JB_ERR_OK if everything went fine, or
1344 * JB_ERR_PARSE if the headers were incomplete.
1346 *********************************************************************/
1347 static jb_err get_server_headers(struct client_state *csp)
1349 int continue_hack_in_da_house = 0;
1352 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1357 * continue hack in da house. Ignore the ending of
1358 * this head and continue enlisting header lines.
1359 * The reason is described below.
1361 enlist(csp->headers, "");
1362 continue_hack_in_da_house = 0;
1365 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1368 * It's a bodyless continue response, don't
1369 * stop header parsing after reaching it's end.
1371 * As a result Privoxy will concatenate the
1372 * next response's head and parse and deliver
1373 * the headers as if they belonged to one request.
1375 * The client will separate them because of the
1376 * empty line between them.
1378 * XXX: What we're doing here is clearly against
1379 * the intended purpose of the continue header,
1380 * and under some conditions (HTTP/1.0 client request)
1381 * it's a standard violation.
1383 * Anyway, "sort of against the spec" is preferable
1384 * to "always getting confused by Continue responses"
1385 * (Privoxy's behaviour before this hack was added)
1387 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1388 continue_hack_in_da_house = 1;
1390 else if (*header == '\0')
1393 * If the header is empty, but the Continue hack
1394 * isn't active, we can assume that we reached the
1395 * end of the buffer before we hit the end of the
1398 * Inform the caller an let it decide how to handle it.
1400 return JB_ERR_PARSE;
1404 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1407 * XXX: Should we quit the request and return a
1408 * out of memory error page instead?
1410 log_error(LOG_LEVEL_ERROR,
1411 "Out of memory while enlisting server headers. %s lost.",
1421 /*********************************************************************
1423 * Function : crunch_reason
1425 * Description : Translates the crunch reason code into a string.
1428 * 1 : rsp = a http_response
1430 * Returns : A string with the crunch reason or an error description.
1432 *********************************************************************/
1433 static const char *crunch_reason(const struct http_response *rsp)
1435 char * reason = NULL;
1437 assert(rsp != NULL);
1440 return "Internal error while searching for crunch reason";
1443 switch (rsp->reason)
1445 case RSP_REASON_UNSUPPORTED:
1446 reason = "Unsupported HTTP feature";
1448 case RSP_REASON_BLOCKED:
1451 case RSP_REASON_UNTRUSTED:
1452 reason = "Untrusted";
1454 case RSP_REASON_REDIRECTED:
1455 reason = "Redirected";
1457 case RSP_REASON_CGI_CALL:
1458 reason = "CGI Call";
1460 case RSP_REASON_NO_SUCH_DOMAIN:
1461 reason = "DNS failure";
1463 case RSP_REASON_FORWARDING_FAILED:
1464 reason = "Forwarding failed";
1466 case RSP_REASON_CONNECT_FAILED:
1467 reason = "Connection failure";
1469 case RSP_REASON_OUT_OF_MEMORY:
1470 reason = "Out of memory (may mask other reasons)";
1473 reason = "No reason recorded";
1481 /*********************************************************************
1483 * Function : send_crunch_response
1485 * Description : Delivers already prepared response for
1486 * intercepted requests, logs the interception
1487 * and frees the response.
1490 * 1 : csp = Current client state (buffers, headers, etc...)
1491 * 1 : rsp = Fully prepared response. Will be freed on exit.
1493 * Returns : Nothing.
1495 *********************************************************************/
1496 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1498 const struct http_request *http = csp->http;
1499 char status_code[4];
1501 assert(rsp != NULL);
1502 assert(rsp->head != NULL);
1507 * Not supposed to happen. If it does
1508 * anyway, treat it as an unknown error.
1510 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1511 /* return code doesn't matter */
1516 /* If rsp is still NULL, we have serious internal problems. */
1517 log_error(LOG_LEVEL_FATAL,
1518 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1522 * Extract the status code from the actual head
1523 * that was send to the client. It is the only
1524 * way to get it right for all requests, including
1525 * the fixed ones for out-of-memory problems.
1527 * A head starts like this: 'HTTP/1.1 200...'
1531 status_code[0] = rsp->head[9];
1532 status_code[1] = rsp->head[10];
1533 status_code[2] = rsp->head[11];
1534 status_code[3] = '\0';
1536 /* Write the answer to the client */
1537 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1538 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1540 /* There is nothing we can do about it. */
1541 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1544 /* Log that the request was crunched and why. */
1545 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1546 http->hostport, http->path, crunch_reason(rsp));
1547 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1548 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1550 /* Clean up and return */
1551 if (cgi_error_memory() != rsp)
1553 free_http_response(rsp);
1560 /*********************************************************************
1562 * Function : request_contains_null_bytes
1564 * Description : Checks for NULL bytes in the request and sends
1565 * an error message to the client if any were found.
1567 * XXX: currently not used, see comment in chat().
1570 * 1 : csp = Current client state (buffers, headers, etc...)
1571 * 2 : buf = Data from the client's request to check.
1572 * 3 : len = The data length.
1574 * Returns : TRUE if the request contained one or more NULL bytes, or
1577 *********************************************************************/
1578 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1580 size_t c_len; /* Request lenght when treated as C string */
1582 c_len = strlen(buf);
1587 * Null byte(s) found. Log the request,
1588 * return an error response and hang up.
1590 size_t tmp_len = c_len;
1595 * Replace NULL byte(s) with '°' characters
1596 * so the request can be logged as string.
1597 * XXX: Is there a better replacement character?
1600 tmp_len += strlen(buf+tmp_len);
1601 } while (tmp_len < len);
1603 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1604 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1605 log_error(LOG_LEVEL_HEADER,
1606 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1608 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1610 /* XXX: Log correct size */
1611 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1621 /*********************************************************************
1623 * Function : crunch_response_triggered
1625 * Description : Checks if the request has to be crunched,
1626 * and delivers the crunch response if necessary.
1629 * 1 : csp = Current client state (buffers, headers, etc...)
1630 * 2 : crunchers = list of cruncher functions to run
1632 * Returns : TRUE if the request was answered with a crunch response
1635 *********************************************************************/
1636 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1638 struct http_response *rsp = NULL;
1639 const struct cruncher *c;
1642 * If CGI request crunching is disabled,
1643 * check the CGI dispatcher out of order to
1644 * prevent unintentional blocks or redirects.
1646 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1647 && (NULL != (rsp = dispatch_cgi(csp))))
1649 /* Deliver, log and free the interception response. */
1650 send_crunch_response(csp, rsp);
1654 for (c = crunchers; c->cruncher != NULL; c++)
1657 * Check the cruncher if either Privoxy is toggled
1658 * on and the request isn't forced, or if the cruncher
1659 * applies to forced requests as well.
1661 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1662 !(csp->flags & CSP_FLAG_FORCED)) ||
1663 (c->flags & CF_IGNORE_FORCE))
1665 rsp = c->cruncher(csp);
1668 /* Deliver, log and free the interception response. */
1669 send_crunch_response(csp, rsp);
1670 #ifdef FEATURE_STATISTICS
1671 if (c->flags & CF_COUNT_AS_REJECT)
1673 csp->flags |= CSP_FLAG_REJECTED;
1675 #endif /* def FEATURE_STATISTICS */
1686 /*********************************************************************
1688 * Function : build_request_line
1690 * Description : Builds the HTTP request line.
1692 * If a HTTP forwarder is used it expects the whole URL,
1693 * web servers only get the path.
1696 * 1 : csp = Current client state (buffers, headers, etc...)
1697 * 2 : fwd = The forwarding spec used for the request
1698 * XXX: Should use http->fwd instead.
1699 * 3 : request_line = The old request line which will be replaced.
1701 * Returns : Nothing. Terminates in case of memory problems.
1703 *********************************************************************/
1704 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1706 struct http_request *http = csp->http;
1708 assert(http->ssl == 0);
1711 * Downgrade http version from 1.1 to 1.0
1712 * if +downgrade action applies.
1714 if ( (csp->action->flags & ACTION_DOWNGRADE)
1715 && (!strcmpic(http->ver, "HTTP/1.1")))
1718 http->ver = strdup("HTTP/1.0");
1720 if (http->ver == NULL)
1722 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1727 * Rebuild the request line.
1729 freez(*request_line);
1730 *request_line = strdup(http->gpc);
1731 string_append(request_line, " ");
1733 if (fwd->forward_host)
1735 string_append(request_line, http->url);
1739 string_append(request_line, http->path);
1741 string_append(request_line, " ");
1742 string_append(request_line, http->ver);
1744 if (*request_line == NULL)
1746 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1748 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1752 /*********************************************************************
1754 * Function : change_request_destination
1756 * Description : Parse a (rewritten) request line and regenerate
1757 * the http request data.
1760 * 1 : csp = Current client state (buffers, headers, etc...)
1762 * Returns : Forwards the parse_http_request() return code.
1763 * Terminates in case of memory problems.
1765 *********************************************************************/
1766 static jb_err change_request_destination(struct client_state *csp)
1768 struct http_request *http = csp->http;
1771 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1772 free_http_request(http);
1773 err = parse_http_request(csp->headers->first->str, http, csp);
1774 if (JB_ERR_OK != err)
1776 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1777 jb_err_to_string(err));
1779 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1780 if (http->ocmd == NULL)
1782 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1789 /*********************************************************************
1791 * Function : get_filter_function
1793 * Description : Decides which content filter function has
1794 * to be applied (if any).
1796 * XXX: Doesn't handle filter_popups()
1797 * because of the different prototype. Probably
1798 * we should ditch filter_popups() anyway, it's
1799 * even less reliable than popup blocking based
1803 * 1 : csp = Current client state (buffers, headers, etc...)
1805 * Returns : The content filter function to run, or
1806 * NULL if no content filter is active
1808 *********************************************************************/
1809 static filter_function_ptr get_filter_function(struct client_state *csp)
1811 filter_function_ptr filter_function = NULL;
1814 * Are we enabling text mode by force?
1816 if (csp->action->flags & ACTION_FORCE_TEXT_MODE)
1819 * Do we really have to?
1821 if (csp->content_type & CT_TEXT)
1823 log_error(LOG_LEVEL_HEADER, "Text mode is already enabled.");
1827 csp->content_type |= CT_TEXT;
1828 log_error(LOG_LEVEL_HEADER, "Text mode enabled by force. Take cover!");
1832 if (!(csp->content_type & CT_DECLARED))
1835 * The server didn't bother to declare a MIME-Type.
1836 * Assume it's text that can be filtered.
1838 * This also regulary happens with 304 responses,
1839 * therefore logging anything here would cause
1842 csp->content_type |= CT_TEXT;
1847 * Choose the applying filter function based on
1848 * the content type and action settings.
1850 if ((csp->content_type & CT_TEXT) &&
1851 (csp->rlist != NULL) &&
1852 (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER])))
1854 filter_function = pcrs_filter_response;
1856 else if ((csp->content_type & CT_GIF) &&
1857 (csp->action->flags & ACTION_DEANIMATE))
1859 filter_function = gif_deanimate_response;
1861 else if ((csp->content_type & CT_JPEG) &&
1862 (csp->action->flags & ACTION_JPEG_INSPECT))
1864 filter_function = jpeg_inspect_response;
1867 return filter_function;
1870 /*********************************************************************
1874 * Description : Once a connection to the client has been accepted,
1875 * this function is called (via serve()) to handle the
1876 * main business of the communication. When this
1877 * function returns, the caller must close the client
1880 * FIXME: chat is nearly thousand lines long.
1884 * 1 : csp = Current client state (buffers, headers, etc...)
1886 * Returns : Nothing.
1888 *********************************************************************/
1889 static void chat(struct client_state *csp)
1891 char buf[BUFFER_SIZE];
1899 int ms_iis5_hack = 0;
1900 size_t byte_count = 0;
1901 int forwarded_connect_retries = 0;
1902 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1903 const struct forward_spec * fwd;
1904 struct http_request *http;
1905 int len; /* for buffer sizes (and negative error codes) */
1907 #ifdef FEATURE_KILL_POPUPS
1908 int block_popups_now = 0; /* bool, 1==currently blocking popups */
1909 #endif /* def FEATURE_KILL_POPUPS */
1911 /* Function that does the content filtering for the current request */
1912 filter_function_ptr content_filter = NULL;
1914 /* Skeleton for HTTP response, if we should intercept the request */
1915 struct http_response *rsp;
1917 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1918 struct list header_list;
1919 struct list *headers = &header_list;
1923 memset(buf, 0, sizeof(buf));
1926 * Read the client's request. Note that since we're not using select() we
1927 * could get blocked here if a client connected, then didn't say anything!
1932 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1934 if (len <= 0) break; /* error! */
1937 * If there is no memory left for buffering the
1938 * request, there is nothing we can do but hang up
1940 if (add_to_iob(csp, buf, len))
1945 req = get_header(csp);
1947 } while ((NULL != req) && ('\0' == *req));
1951 /* Request received. Validate and parse it. */
1955 * XXX: Temporary disabled to prevent problems
1956 * with POST requests whose bodies are allowed to
1957 * contain NULL bytes. BR#1730105.
1959 * The main purpose of this check is to properly
1960 * log stuff like BitTorrent traffic and other junk
1961 * that hits public proxies. It's not required for
1962 * Privoxy to functions as those requests are discarded
1965 * It probably should be rewritten to only check
1966 * the head of the request. Another option would
1967 * be to let all POST requests pass, although that
1968 * may not be good enough.
1970 if (request_contains_null_bytes(csp, buf, len))
1972 /* NULL bytes found and dealt with, just hang up. */
1977 /* Does the request line look invalid? */
1978 if (client_protocol_is_unsupported(csp, req))
1981 * Yes. The request has already been
1982 * answered with a error response, the buffers
1983 * were freed and we're done with chatting.
1988 #ifdef FEATURE_FORCE_LOAD
1990 * If this request contains the FORCE_PREFIX and blocks
1991 * aren't enforced, get rid of it and set the force flag.
1993 if (strstr(req, FORCE_PREFIX))
1995 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1997 log_error(LOG_LEVEL_FORCE,
1998 "Ignored force prefix in request: \"%s\".", req);
2002 strclean(req, FORCE_PREFIX);
2003 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2004 csp->flags |= CSP_FLAG_FORCED;
2008 #endif /* def FEATURE_FORCE_LOAD */
2009 err = parse_http_request(req, http, csp);
2010 if (JB_ERR_OK != err)
2012 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2018 if (http->cmd == NULL)
2020 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2021 /* XXX: Use correct size */
2022 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2023 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2025 free_http_request(http);
2029 /* grab the rest of the client's headers */
2033 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
2035 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2038 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2043 * If there is no memory left for buffering the
2044 * request, there is nothing we can do but hang up
2046 if (add_to_iob(csp, buf, len))
2053 if (p == NULL) break;
2060 if (http->host == NULL)
2063 * If we still don't know the request destination,
2064 * the request is invalid or the client uses
2065 * Privoxy without its knowledge.
2067 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2070 * Our attempts to get the request destination
2071 * elsewhere failed or Privoxy is configured
2072 * to only accept proxy requests.
2074 * An error response has already been send
2075 * and we're done here.
2082 * Determine the actions for this URL
2084 #ifdef FEATURE_TOGGLE
2085 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2087 /* Most compatible set of actions (i.e. none) */
2088 init_current_action(csp->action);
2091 #endif /* ndef FEATURE_TOGGLE */
2093 url_actions(http, csp);
2097 * Save a copy of the original request for logging
2099 http->ocmd = strdup(http->cmd);
2101 if (http->ocmd == NULL)
2103 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2106 enlist(csp->headers, http->cmd);
2108 /* Append the previously read headers */
2109 list_append_list_unique(csp->headers, headers);
2110 destroy_list(headers);
2113 * If the user has not supplied any wafers, and the user has not
2114 * told us to suppress the vanilla wafer, then send the vanilla wafer.
2116 if (list_is_empty(csp->action->multi[ACTION_MULTI_WAFER])
2117 && ((csp->action->flags & ACTION_VANILLA_WAFER) != 0))
2119 enlist(csp->action->multi[ACTION_MULTI_WAFER], VANILLA_WAFER);
2122 err = sed(client_patterns, add_client_headers, csp);
2123 if (JB_ERR_OK != err)
2125 assert(err == JB_ERR_PARSE);
2126 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2128 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2130 if (strcmp(http->cmd, csp->headers->first->str))
2133 * A header filter rewrote the request line,
2134 * modify the http request accordingly.
2136 if (JB_ERR_OK != change_request_destination(csp))
2138 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2139 /* XXX: Use correct size */
2140 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2141 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2143 free_http_request(http);
2148 /* decide how to route the HTTP request */
2149 if (NULL == (fwd = forward_url(http, csp)))
2151 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2152 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2155 /* build the http request to send to the server
2156 * we have to do one of the following:
2158 * create = use the original HTTP request to create a new
2159 * HTTP request that has either the path component
2160 * without the http://domainspec (w/path) or the
2161 * full orininal URL (w/url)
2162 * Note that the path and/or the HTTP version may
2163 * have been altered by now.
2165 * connect = Open a socket to the host:port of the server
2166 * and short-circuit server and client socket.
2168 * pass = Pass the request unchanged if forwarding a CONNECT
2169 * request to a parent proxy. Note that we'll be sending
2170 * the CFAIL message ourselves if connecting to the parent
2171 * fails, but we won't send a CSUCCEED message if it works,
2172 * since that would result in a double message (ours and the
2173 * parent's). After sending the request to the parent, we simply
2176 * here's the matrix:
2179 * +--------+--------+
2181 * 0 | create | connect|
2183 * Forwarding +--------+--------+
2185 * 1 | create | pass |
2187 * +--------+--------+
2192 * Check if a CONNECT request is allowable:
2193 * In the absence of a +limit-connect action, allow only port 443.
2194 * If there is an action, allow whatever matches the specificaton.
2198 if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443)
2199 || (csp->action->flags & ACTION_LIMIT_CONNECT
2200 && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) )
2202 if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS)
2205 * The response may confuse some clients,
2206 * but makes unblocking easier.
2208 * XXX: It seems to work with all major browsers,
2209 * so we should consider returning a body by default someday ...
2211 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2212 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2213 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2215 csp->action->flags |= ACTION_BLOCK;
2220 write_socket(csp->cfd, CFORBIDDEN, strlen(CFORBIDDEN));
2221 log_error(LOG_LEVEL_INFO, "Request from %s denied. "
2222 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2223 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2225 assert(NULL != csp->http->ocmd);
2226 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 403 0", csp->ip_addr_str, csp->http->ocmd);
2228 list_remove_all(csp->headers);
2230 * XXX: For consistency we might want to log a crunch message here.
2239 freez(csp->headers->first->str);
2240 build_request_line(csp, fwd, &csp->headers->first->str);
2243 hdr = list_to_text(csp->headers);
2246 /* FIXME Should handle error properly */
2247 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2251 * We have a request. Check if one of the crunchers wants it.
2253 if (crunch_response_triggered(csp, crunchers_all))
2256 * Yes. The client got the crunch response
2257 * and we are done here after cleaning up.
2260 list_remove_all(csp->headers);
2266 * The headers can't be removed earlier because
2267 * they were still needed for the referrer check
2268 * in case of CGI crunches.
2270 * XXX: Would it be worth to move the referrer check
2271 * into client_referrer() and set a flag if it's trusted?
2273 list_remove_all(csp->headers);
2275 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2277 if (fwd->forward_host)
2279 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2280 fwd->forward_host, fwd->forward_port, http->hostport);
2284 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2287 /* here we connect to the server, gateway, or the forwarder */
2289 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2290 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2292 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2293 forwarded_connect_retries, http->hostport);
2296 if (csp->sfd == JB_INVALID_SOCKET)
2298 if (fwd->type != SOCKS_NONE)
2301 rsp = error_response(csp, "forwarding-failed", errno);
2303 else if (errno == EINVAL)
2305 rsp = error_response(csp, "no-such-domain", errno);
2309 rsp = error_response(csp, "connect-failed", errno);
2310 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2315 /* Write the answer to the client */
2318 send_crunch_response(csp, rsp);
2325 if (fwd->forward_host || (http->ssl == 0))
2327 /* write the client's (modified) header to the server
2328 * (along with anything else that may be in the buffer)
2331 if (write_socket(csp->sfd, hdr, strlen(hdr))
2332 || (flush_socket(csp->sfd, csp) < 0))
2334 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2337 rsp = error_response(csp, "connect-failed", errno);
2341 send_crunch_response(csp, rsp);
2351 * We're running an SSL tunnel and we're not forwarding,
2352 * so just send the "connect succeeded" message to the
2353 * client, flush the rest, and get out of the way.
2355 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2363 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2365 /* we're finished with the client's header */
2368 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2370 /* pass data between the client and server
2371 * until one or the other shuts down the connection.
2380 * FD_ZERO here seems to point to an errant macro which crashes.
2381 * So do this by hand for now...
2383 memset(&rfds,0x00,sizeof(fd_set));
2387 FD_SET(csp->cfd, &rfds);
2388 FD_SET(csp->sfd, &rfds);
2390 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2394 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2398 /* this is the body of the browser's request
2399 * just read it and write it.
2402 if (FD_ISSET(csp->cfd, &rfds))
2404 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2408 break; /* "game over, man" */
2411 if (write_socket(csp->sfd, buf, (size_t)len))
2413 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2420 * The server wants to talk. It could be the header or the body.
2421 * If `hdr' is null, then it's the header otherwise it's the body.
2422 * FIXME: Does `hdr' really mean `host'? No.
2426 if (FD_ISSET(csp->sfd, &rfds))
2429 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2433 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2435 if (http->ssl && (fwd->forward_host == NULL))
2438 * Just hang up. We already confirmed the client's CONNECT
2439 * request with status code 200 and unencrypted content is
2440 * no longer welcome.
2442 log_error(LOG_LEVEL_ERROR,
2443 "CONNECT already confirmed. Unable to tell the client about the problem.");
2447 rsp = error_response(csp, "connect-failed", errno);
2451 send_crunch_response(csp, rsp);
2457 /* Add a trailing zero. This lets filter_popups
2458 * use string operations.
2462 #ifdef FEATURE_KILL_POPUPS
2463 /* Filter the popups on this read. */
2464 if (block_popups_now)
2466 filter_popups(buf, csp);
2468 #endif /* def FEATURE_KILL_POPUPS */
2470 /* Normally, this would indicate that we've read
2471 * as much as the server has sent us and we can
2472 * close the client connection. However, Microsoft
2473 * in its wisdom has released IIS/5 with a bug that
2474 * prevents it from sending the trailing \r\n in
2475 * a 302 redirect header (and possibly other headers).
2476 * To work around this if we've haven't parsed
2477 * a full header we'll append a trailing \r\n
2478 * and see if this now generates a valid one.
2480 * This hack shouldn't have any impacts. If we've
2481 * already transmitted the header or if this is a
2482 * SSL connection, then we won't bother with this
2483 * hack. So we only work on partially received
2484 * headers. If we append a \r\n and this still
2485 * doesn't generate a valid header, then we won't
2486 * transmit anything to the client.
2491 if (server_body || http->ssl)
2494 * If we have been buffering up the document,
2495 * now is the time to apply content modification
2496 * and send the result to the client.
2501 * If the content filter fails, use the original
2502 * buffer and length.
2503 * (see p != NULL ? p : csp->iob->cur below)
2505 if (NULL == (p = (*content_filter)(csp)))
2507 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2510 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2512 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2515 hdr = list_to_text(csp->headers);
2518 /* FIXME Should handle error properly */
2519 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2523 * Shouldn't happen because this was the second sed run
2524 * and tags are only created for the first one.
2526 assert(!crunch_response_triggered(csp, crunchers_all));
2528 if (write_socket(csp->cfd, hdr, strlen(hdr))
2529 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2531 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2541 break; /* "game over, man" */
2545 * This is NOT the body, so
2546 * Let's pretend the server just sent us a blank line.
2548 snprintf(buf, sizeof(buf), "\r\n");
2549 len = (int)strlen(buf);
2552 * Now, let the normal header parsing algorithm below do its
2553 * job. If it fails, we'll exit instead of continuing.
2560 * If this is an SSL connection or we're in the body
2561 * of the server document, just write it to the client,
2562 * unless we need to buffer the body for later content-filtering
2565 if (server_body || http->ssl)
2570 * If there is no memory left for buffering the content, or the buffer limit
2571 * has been reached, switch to non-filtering mode, i.e. make & write the
2572 * header, flush the iob and buf, and get out of the way.
2574 if (add_to_iob(csp, buf, len))
2579 log_error(LOG_LEVEL_ERROR, "Flushing header and buffers. Stepping back from filtering.");
2580 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2582 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2584 hdr = list_to_text(csp->headers);
2588 * Memory is too tight to even generate the header.
2589 * Send our static "Out-of-memory" page.
2591 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2592 rsp = cgi_error_memory();
2593 send_crunch_response(csp, rsp);
2598 if (crunch_response_triggered(csp, crunchers_light))
2601 * One of the tags created by a server-header
2602 * tagger triggered a crunch. We already
2603 * delivered the crunch response to the client
2604 * and are done here after cleaning up.
2610 hdrlen = strlen(hdr);
2612 if (write_socket(csp->cfd, hdr, hdrlen)
2613 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
2614 || (write_socket(csp->cfd, buf, (size_t)len)))
2616 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2622 byte_count += hdrlen + (size_t)flushed + (size_t)len;
2624 content_filter = NULL;
2631 if (write_socket(csp->cfd, buf, (size_t)len))
2633 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2637 byte_count += (size_t)len;
2642 /* we're still looking for the end of the
2643 * server's header ... (does that make header
2644 * parsing an "out of body experience" ?
2648 * buffer up the data we just read. If that fails,
2649 * there's little we can do but send our static
2650 * out-of-memory page.
2652 if (add_to_iob(csp, buf, len))
2654 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2655 rsp = cgi_error_memory();
2656 send_crunch_response(csp, rsp);
2661 /* Convert iob into something sed() can digest */
2662 if (JB_ERR_PARSE == get_server_headers(csp))
2666 /* Well, we tried our MS IIS/5
2667 * hack and it didn't work.
2668 * The header is incomplete
2669 * and there isn't anything
2670 * we can do about it.
2676 /* Since we have to wait for
2677 * more from the server before
2678 * we can parse the headers
2679 * we just continue here.
2685 /* Did we actually get anything? */
2686 if (NULL == csp->headers->first)
2688 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2689 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2690 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2691 free_http_request(http);
2695 /* we have now received the entire header.
2696 * filter it and send the result to the client
2698 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2700 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2702 hdr = list_to_text(csp->headers);
2705 /* FIXME Should handle error properly */
2706 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2709 if (crunch_response_triggered(csp, crunchers_light))
2712 * One of the tags created by a server-header
2713 * tagger triggered a crunch. We already
2714 * delivered the crunch response to the client
2715 * and are done here after cleaning up.
2720 /* Buffer and pcrs filter this if appropriate. */
2722 if (!http->ssl) /* We talk plaintext */
2725 #ifdef FEATURE_KILL_POPUPS
2726 /* Start blocking popups if appropriate. */
2727 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2728 (csp->action->flags & ACTION_NO_POPUPS) != 0) /* Policy allows */
2730 block_popups_now = 1;
2732 * Filter the part of the body that came in the same read
2733 * as the last headers:
2735 filter_popups(csp->iob->cur, csp);
2737 #endif /* def FEATURE_KILL_POPUPS */
2738 content_filter = get_filter_function(csp);
2741 * Only write if we're not buffering for content modification
2743 if (!content_filter)
2745 /* write the server's (modified) header to
2746 * the client (along with anything else that
2747 * may be in the buffer)
2750 if (write_socket(csp->cfd, hdr, strlen(hdr))
2751 || ((len = flush_socket(csp->cfd, csp)) < 0))
2753 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2755 /* the write failed, so don't bother
2756 * mentioning it to the client...
2757 * it probably can't hear us anyway.
2763 byte_count += (size_t)len;
2766 /* we're finished with the server's header */
2771 /* If this was a MS IIS/5 hack then it means
2772 * the server has already closed the
2773 * connection. Nothing more to read. Time
2784 return; /* huh? we should never get here */
2787 if (csp->content_length == 0)
2790 * If Privoxy didn't recalculate the
2791 * Content-Lenght, byte_count is still
2794 csp->content_length = byte_count;
2797 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2798 csp->ip_addr_str, http->ocmd, csp->content_length);
2802 /*********************************************************************
2806 * Description : This is little more than chat. We only "serve" to
2807 * to close any socket that chat may have opened.
2810 * 1 : csp = Current client state (buffers, headers, etc...)
2814 *********************************************************************/
2816 void serve(struct client_state *csp)
2817 #else /* ifndef AMIGA */
2818 static void serve(struct client_state *csp)
2819 #endif /* def AMIGA */
2822 close_socket(csp->cfd);
2824 if (csp->sfd != JB_INVALID_SOCKET)
2826 close_socket(csp->sfd);
2829 csp->flags &= ~CSP_FLAG_ACTIVE;
2835 /*********************************************************************
2837 * Function : server_thread
2839 * Description : We only exist to call `serve' in a threaded environment.
2842 * 1 : data = Current client state (buffers, headers, etc...)
2844 * Returns : Always 0.
2846 *********************************************************************/
2847 static int32 server_thread(void *data)
2849 serve((struct client_state *) data);
2856 /*********************************************************************
2860 * Description : Print usage info & exit.
2862 * Parameters : Pointer to argv[0] for identifying ourselves
2866 *********************************************************************/
2867 static void usage(const char *myname)
2869 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2873 #endif /* defined(unix) */
2876 "[--no-daemon] [--pidfile pidfile] [--user user[.group]] "
2877 #endif /* defined(unix) */
2878 "[--version] [configfile]\n"
2879 "Aborting\n", myname);
2886 /*********************************************************************
2888 * Function : initialize_mutexes
2890 * Description : Prepares mutexes if mutex support is available.
2894 * Returns : Void, exits in case of errors.
2896 *********************************************************************/
2897 static void initialize_mutexes(void)
2901 #ifdef FEATURE_PTHREAD
2903 * Prepare global mutex semaphores
2905 err = pthread_mutex_init(&log_mutex, 0);
2907 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2910 * XXX: The assumptions below are a bit naive
2911 * and can cause locks that aren't necessary.
2913 * For example older FreeBSD versions (< 6.x?)
2914 * have no gethostbyname_r, but gethostbyname is
2917 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2918 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2919 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2921 * XXX: should we use a single mutex for
2922 * localtime() and gmtime() as well?
2924 #ifndef HAVE_GMTIME_R
2925 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2926 #endif /* ndef HAVE_GMTIME_R */
2928 #ifndef HAVE_LOCALTIME_R
2929 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2930 #endif /* ndef HAVE_GMTIME_R */
2933 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2934 #endif /* ndef HAVE_RANDOM */
2935 #endif /* FEATURE_PTHREAD */
2938 * TODO: mutex support for mingw32 would be swell.
2943 printf("Fatal error. Mutex initialization failed: %s.\n",
2952 /*********************************************************************
2956 * Description : Load the config file and start the listen loop.
2957 * This function is a lot more *sane* with the `load_config'
2958 * and `listen_loop' functions; although it stills does
2959 * a *little* too much for my taste.
2962 * 1 : argc = Number of parameters (including $0).
2963 * 2 : argv = Array of (char *)'s to the parameters.
2965 * Returns : 1 if : can't open config file, unrecognized directive,
2966 * stats requested in multi-thread mode, can't open the
2967 * log file, can't open the jar file, listen port is invalid,
2968 * any load fails, and can't bind port.
2970 * Else main never returns, the process must be signaled
2971 * to terminate execution. Or, on Windows, use the
2972 * "File", "Exit" menu option.
2974 *********************************************************************/
2976 int real_main(int argc, const char *argv[])
2978 int main(int argc, const char *argv[])
2982 unsigned int random_seed;
2984 struct passwd *pw = NULL;
2985 struct group *grp = NULL;
2994 #if !defined(_WIN32)
3002 * Parse the command line arguments
3004 * XXX: simply printing usage information in case of
3005 * invalid arguments isn't particular user friendly.
3007 while (++argc_pos < argc)
3010 /* Check to see if the service must be installed or uninstalled */
3011 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3013 const char *pName = argv[argc_pos] + 9;
3016 exit( (install_service(pName)) ? 0 : 1 );
3018 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3020 const char *pName = argv[argc_pos] + 11;
3023 exit((uninstall_service(pName)) ? 0 : 1);
3025 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3027 bRunAsService = TRUE;
3028 w32_set_service_cwd();
3029 atexit(w32_service_exit_notify);
3032 #endif /* defined(_WIN32) */
3035 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3037 if (strcmp(argv[argc_pos], "--help") == 0)
3042 else if(strcmp(argv[argc_pos], "--version") == 0)
3044 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3050 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3055 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3057 if (++argc_pos == argc) usage(argv[0]);
3058 pidfile = strdup(argv[argc_pos]);
3061 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3063 if (++argc_pos == argc) usage(argv[argc_pos]);
3065 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3068 if (NULL == (grp = getgrnam(p)))
3070 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3074 if (NULL == (pw = getpwnam(argv[argc_pos])))
3076 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3079 if (p != NULL) *--p = '\0';
3082 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3086 #endif /* defined(unix) */
3088 else if (argc_pos + 1 != argc)
3091 * This is neither the last command line
3092 * option, nor was it recognized before,
3093 * therefore it must be invalid.
3099 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3101 configfile = argv[argc_pos];
3104 } /* -END- while (more arguments) */
3107 if ( *configfile != '/' )
3109 char cwd[BUFFER_SIZE];
3111 size_t abs_file_size;
3113 /* make config-filename absolute here */
3114 if (NULL == getcwd(cwd, sizeof(cwd)))
3116 perror("failed to get current working directory");
3121 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3122 basedir = strdup(cwd);
3124 if (NULL == basedir ||
3125 NULL == (abs_file = malloc(abs_file_size)))
3127 perror("malloc failed");
3130 strlcpy(abs_file, basedir, abs_file_size);
3131 strlcat(abs_file, "/", abs_file_size );
3132 strlcat(abs_file, configfile, abs_file_size);
3133 configfile = abs_file;
3135 #endif /* defined unix */
3139 clients->next = NULL;
3143 #elif defined(_WIN32)
3147 /* Prepare mutexes if supported and necessary. */
3148 initialize_mutexes();
3150 random_seed = (unsigned int)time(NULL);
3152 srandom(random_seed);
3155 #endif /* ifdef HAVE_RANDOM */
3158 * Unix signal handling
3160 * Catch the abort, interrupt and terminate signals for a graceful exit
3161 * Catch the hangup signal so the errlog can be reopened.
3162 * Ignore the broken pipe signals (FIXME: Why?)
3164 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3167 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3168 const int ignored_signals[] = { SIGPIPE, 0 };
3170 for (idx = 0; catched_signals[idx] != 0; idx++)
3172 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3173 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3175 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3176 #endif /* ifdef sun */
3178 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3182 for (idx = 0; ignored_signals[idx] != 0; idx++)
3184 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3186 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3191 #else /* ifdef _WIN32 */
3192 # ifdef _WIN_CONSOLE
3194 * We *are* in a windows console app.
3195 * Print a verbose messages about FAQ's and such
3197 printf("%s", win32_blurb);
3198 # endif /* def _WIN_CONSOLE */
3199 #endif /* def _WIN32 */
3202 /* Initialize the CGI subsystem */
3203 cgi_init_error_messages();
3206 * If runnig on unix and without the --nodaemon
3207 * option, become a daemon. I.e. fork, detach
3208 * from tty and get process group leadership
3221 if ( pid < 0 ) /* error */
3226 else if ( pid != 0 ) /* parent */
3231 * must check for errors
3232 * child died due to missing files aso
3235 wpid = waitpid( pid, &status, WNOHANG );
3244 /* Should be more portable, but not as well tested */
3249 #else /* ndef __FreeBSD__ */
3251 #endif /* ndef __FreeBSD__ */
3252 fd = open("/dev/tty", O_RDONLY);
3255 /* no error check here */
3256 ioctl( fd, TIOCNOTTY,0 );
3260 /* FIXME: should close stderr (fd 2) here too, but the test
3262 * and load config file is done in listen_loop() and puts
3263 * some messages on stderr there.
3270 } /* -END- if (!no_daemon) */
3273 * As soon as we have written the PID file, we can switch
3274 * to the user and group ID indicated by the --user option
3280 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3282 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3288 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3290 if (chroot(pw->pw_dir) < 0)
3292 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3296 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3299 if (setuid(pw->pw_uid))
3301 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3305 char putenv_dummy[64];
3307 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3308 if (putenv(putenv_dummy) != 0)
3310 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3313 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3314 if (putenv(putenv_dummy) != 0)
3316 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3322 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3325 #endif /* defined unix */
3328 /* This will be FALSE unless the command line specified --service
3332 /* Yup, so now we must attempt to establish a connection
3333 * with the service dispatcher. This will only work if this
3334 * process was launched by the service control manager to
3335 * actually run as a service. If this isn't the case, i've
3336 * known it take around 30 seconds or so for the call to return.
3339 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3340 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3342 /* Service has run, and at this point is now being stopped, so just return */
3347 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3349 /* An error occurred. Usually it's because --service was wrongly specified
3350 * and we were unable to connect to the Service Control Dispatcher because
3351 * it wasn't expecting us and is therefore not listening.
3353 * For now, just continue below to call the listen_loop function.
3356 #endif /* def _WIN32 */
3366 /*********************************************************************
3368 * Function : bind_port_helper
3370 * Description : Bind the listen port. Handles logging, and aborts
3374 * 1 : config = Privoxy configuration. Specifies port
3377 * Returns : Port that was opened.
3379 *********************************************************************/
3380 static jb_socket bind_port_helper(struct configuration_spec * config)
3385 if ( (config->haddr != NULL)
3386 && (config->haddr[0] == '1')
3387 && (config->haddr[1] == '2')
3388 && (config->haddr[2] == '7')
3389 && (config->haddr[3] == '.') )
3391 log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only",
3394 else if (config->haddr == NULL)
3396 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3401 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3402 config->hport, config->haddr);
3405 result = bind_port(config->haddr, config->hport, &bfd);
3412 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3413 "There may be another Privoxy or some other "
3414 "proxy running on port %d",
3415 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3416 config->hport, config->hport);
3419 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3420 "The hostname is not resolvable",
3421 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3424 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3425 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3428 /* shouldn't get here */
3429 return JB_INVALID_SOCKET;
3432 config->need_bind = 0;
3439 /* Without this simple workaround we get this compiler warning from _beginthread
3440 * warning C4028: formal parameter 1 different from declaration
3442 void w32_service_listen_loop(void *p)
3446 #endif /* def _WIN32 */
3449 /*********************************************************************
3451 * Function : listen_loop
3453 * Description : bind the listen port and enter a "FOREVER" listening loop.
3459 *********************************************************************/
3460 static void listen_loop(void)
3462 struct client_state *csp = NULL;
3464 struct configuration_spec * config;
3466 config = load_config();
3468 bfd = bind_port_helper(config);
3470 #ifdef FEATURE_GRACEFUL_TERMINATION
3471 while (!g_terminate)
3476 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3477 while (waitpid(-1, NULL, WNOHANG) > 0)
3479 /* zombie children */
3481 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3484 * Free data that was used by died threads
3490 * Re-open the errlog after HUP signal
3492 if (received_hup_signal)
3494 init_error_log(Argv[0], config->logfile, config->debug);
3495 received_hup_signal = 0;
3500 #ifdef FEATURE_COOKIE_JAR
3502 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3503 * already open. I think unload_configfile() is not being run, which should do
3504 * this work. Until that can get resolved, we'll use this workaround.
3508 if (csp->config->jar)
3510 fclose(csp->config->jar);
3511 csp->config->jar = NULL;
3513 #endif /* FEATURE_COOKIE_JAR */
3514 #endif /* __OS2__ */
3516 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3518 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3522 csp->flags |= CSP_FLAG_ACTIVE;
3523 csp->sfd = JB_INVALID_SOCKET;
3525 csp->config = config = load_config();
3527 if ( config->need_bind )
3530 * Since we were listening to the "old port", we will not see
3531 * a "listen" param change until the next IJB request. So, at
3532 * least 1 more request must be made for us to find the new
3533 * setting. I am simply closing the old socket and binding the
3536 * Which-ever is correct, we will serve 1 more page via the
3537 * old settings. This should probably be a "show-proxy-args"
3538 * request. This should not be a so common of an operation
3539 * that this will hurt people's feelings.
3544 bfd = bind_port_helper(config);
3547 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3549 if (!accept_connection(csp, bfd))
3551 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3564 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3567 #ifdef FEATURE_TOGGLE
3568 if (global_toggle_state)
3570 csp->flags |= CSP_FLAG_TOGGLED_ON;
3572 #endif /* def FEATURE_TOGGLE */
3574 if (run_loader(csp))
3576 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3577 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3581 if (block_acl(NULL,csp))
3583 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3584 close_socket(csp->cfd);
3588 #endif /* def FEATURE_ACL */
3590 /* add it to the list of clients */
3591 csp->next = clients->next;
3592 clients->next = csp;
3594 if (config->multi_threaded)
3598 /* this is a switch () statment in the C preprocessor - ugh */
3599 #undef SELECTED_ONE_OPTION
3601 /* Use Pthreads in preference to native code */
3602 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3603 #define SELECTED_ONE_OPTION
3605 pthread_t the_thread;
3606 pthread_attr_t attrs;
3608 pthread_attr_init(&attrs);
3609 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3610 errno = pthread_create(&the_thread, &attrs,
3611 (void * (*)(void *))serve, csp);
3612 child_id = errno ? -1 : 0;
3613 pthread_attr_destroy(&attrs);
3617 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3618 #define SELECTED_ONE_OPTION
3619 child_id = _beginthread(
3620 (void (*)(void *))serve,
3625 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3626 #define SELECTED_ONE_OPTION
3627 child_id = _beginthread(
3628 (void(* _Optlink)(void*))serve,
3634 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3635 #define SELECTED_ONE_OPTION
3637 thread_id tid = spawn_thread
3638 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3640 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3642 child_id = (int) tid;
3651 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3652 #define SELECTED_ONE_OPTION
3653 csp->cfd = ReleaseSocket(csp->cfd, -1);
3656 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3657 NP_Output, Output(),
3658 NP_CloseOutput, FALSE,
3659 NP_Name, (ULONG)"privoxy child",
3663 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3664 NP_Output, Output(),
3665 NP_CloseOutput, FALSE,
3666 NP_Name, (ULONG)"privoxy child",
3667 NP_StackSize, 200*1024,
3673 ((struct Task *)child_id)->tc_UserData = csp;
3674 Signal((struct Task *)child_id, SIGF_SINGLE);
3679 #if !defined(SELECTED_ONE_OPTION)
3682 /* This block is only needed when using fork().
3683 * When using threads, the server thread was
3684 * created and run by the call to _beginthread().
3686 if (child_id == 0) /* child */
3689 #ifdef FEATURE_TOGGLE
3690 int inherited_toggle_state = global_toggle_state;
3691 #endif /* def FEATURE_TOGGLE */
3696 * If we've been toggled or we've blocked the request, tell Mom
3699 #ifdef FEATURE_TOGGLE
3700 if (inherited_toggle_state != global_toggle_state)
3702 rc |= RC_FLAG_TOGGLED;
3704 #endif /* def FEATURE_TOGGLE */
3706 #ifdef FEATURE_STATISTICS
3707 if (csp->flags & CSP_FLAG_REJECTED)
3709 rc |= RC_FLAG_BLOCKED;
3711 #endif /* ndef FEATURE_STATISTICS */
3715 else if (child_id > 0) /* parent */
3717 /* in a fork()'d environment, the parent's
3718 * copy of the client socket and the CSP
3722 #if !defined(_WIN32) && !defined(__CYGWIN__)
3724 wait( &child_status );
3727 * Evaluate child's return code: If the child has
3728 * - been toggled, toggle ourselves
3729 * - blocked its request, bump up the stats counter
3732 #ifdef FEATURE_TOGGLE
3733 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3735 global_toggle_state = !global_toggle_state;
3737 #endif /* def FEATURE_TOGGLE */
3739 #ifdef FEATURE_STATISTICS
3741 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3745 #endif /* def FEATURE_STATISTICS */
3747 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3748 close_socket(csp->cfd);
3749 csp->flags &= ~CSP_FLAG_ACTIVE;
3753 #undef SELECTED_ONE_OPTION
3754 /* end of cpp switch () */
3756 if (child_id < 0) /* failed */
3758 char buf[BUFFER_SIZE];
3760 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3762 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3764 write_socket(csp->cfd, buf, strlen(buf));
3765 close_socket(csp->cfd);
3766 csp->flags &= ~CSP_FLAG_ACTIVE;
3777 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3779 /* Clean up. Aim: free all memory (no leaks) */
3780 #ifdef FEATURE_GRACEFUL_TERMINATION
3782 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3784 unload_current_config_file();
3785 unload_current_actions_file();
3786 unload_current_re_filterfile();
3787 #ifdef FEATURE_TRUST
3788 unload_current_trust_file();
3791 if (config->multi_threaded)
3798 } while ((clients->next != NULL) && (--i > 0));
3802 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3813 #ifdef FEATURE_COOKIE_JAR
3814 if (NULL != config->jar)
3816 fclose(config->jar);
3820 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3821 /* Cleanup - remove taskbar icon etc. */
3826 #endif /* FEATURE_GRACEFUL_TERMINATION */