1 const char jcc_rcs[] = "$Id: jcc.c,v 1.124 2007/02/23 14:59:54 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 1.124 2007/02/23 14:59:54 fabiankeil
37 * Speed up NULL byte escaping and only log the complete
38 * NULL byte requests with header debugging enabled.
40 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
41 * Answer requests that contain NULL bytes with
42 * a custom response instead of waiting for more
43 * data until the client eventually hangs up.
45 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
46 * - Move delivery and logging of crunched responses
47 * from chat() into send_crunch_response().
48 * - Display the reason for generating http_responses.
49 * - Log the content length for LOG_LEVEL_CLF correctly
50 * (still incorrect for some fixed responses).
51 * - Reword an incorrect comment about
52 * treat-forbidden-connects-like-blocks violating
54 * - Add some log messages.
56 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
57 * Move mutex initialization into separate
58 * function and exit in case of errors.
60 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
61 * - Start to reduce chat()'s line count and move
62 * parts of it into separate functions.
63 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
65 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
66 * - Add Proxy-Agent header to HTTP snippets that are
67 * supposed to reach HTTP clients only.
68 * - Made a few CONNECT log messages more descriptive.
69 * - Catch completely empty server responses (as seen
70 * with Tor's fake ".noconnect" top level domain).
71 * - Use shiny new "forwarding-failed" template for socks errors.
73 * Revision 1.118 2007/01/07 07:43:43 joergs
74 * AmigaOS4 support added.
76 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
77 * Added config option accept-intercepted-requests
78 * and disabled it by default.
80 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
81 * Reverted parts of my last commit
82 * to keep error handling working.
84 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
85 * Fixed gcc43 conversion warnings.
87 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
88 * Fix -pedantic ISO C warning about converting
89 * from function pointer to object pointer.
91 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
92 * Silence compiler warning I introduced with my last commit.
94 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
95 * Mutex protect rand() if POSIX threading
96 * is used, warn the user if that's not possible
97 * and stop using it on _WIN32 where it could
100 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
101 * Don't prevent core dumps by catching SIGABRT.
102 * It's rude and makes debugging unreasonable painful.
104 * Revision 1.110 2006/12/13 14:52:53 etresoft
105 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
107 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
108 * Privoxy is now able to run as intercepting
109 * proxy in combination with any packet filter
110 * that does the port redirection. The destination
111 * is extracted from the "Host:" header which
112 * should be available for nearly all requests.
114 * Moved HTTP snipplets into jcc.c.
115 * Added error message for gopher proxy requests.
117 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
118 * Only unlink the pidfile if it's actually used.
120 * Change order of interception checks to make
121 * it possible to block or redirect requests for
124 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
125 * Make pthread mutex locking more generic. Instead of
126 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
127 * and use mutex locking unless there is an _r function
128 * available. Better safe than sorry.
130 * Fixes "./configure --disable-pthread" and should result
131 * in less threading-related problems on pthread-using platforms,
132 * but it still doesn't fix BR#1122404.
134 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
135 * Move pthread.h inclusion from jcc.c to jcc.h.
136 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
138 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
139 * Don't exit after receiving the second SIGHUP on Solaris.
141 * Fixes BR 1052235, but the same problem may exist on other
142 * systems. Once 3.0.6 is out we should use sigset()
143 * where available and see if it breaks anything.
145 * Revision 1.104 2006/09/23 13:26:38 roro
146 * Replace TABs by spaces in source code.
148 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
149 * Fix +redirect{}. Didn't work with -fast-redirects.
151 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
152 * Respond with 400 and a short text message
153 * if the client tries to use Privoxy as FTP proxy.
155 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
156 * Make number of retries in case of forwarded-connect problems
157 * a config file option (forwarded-connect-retries) and use 0 as
160 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
161 * Set random(3) seed.
163 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
164 * Follow the OpenBSD port's lead and protect the resolve
165 * functions on OpenBSD as well.
167 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
168 * --user fix. Only use the user as group if no group is specified.
169 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
171 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
172 * Windows service (re-)integration
176 * --install[:service_name]
177 * --uninstall[:service_name]
180 * They work as follows:
181 * --install will create a service for you and then terminate.
182 * By default the service name will be "privoxy" (without the quotes).
183 * However you can run multiple services if you wish, just by adding
184 * a colon and then a name (no spaces).
186 * --uninstall follows the exact same rules a --install.
188 * --service is used when the program is executed by the service
189 * control manager, and in normal circumstances would never be
190 * used as a command line argument.
192 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
193 * Windows service integration
195 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
196 * Incorporate Fabian Keil's patch work:
197 http://www.fabiankeil.de/sourcecode/privoxy/
199 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
200 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
201 * with what was really the latest development (the v_3_0_branch branch)
203 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
204 * Thanks to Jindrich Makovicka for a race condition fix for the log
205 * file. The race condition remains for non-pthread implementations.
206 * Reference patch #1175720.
208 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
209 * Add the ability to check jpeg images for invalid
210 * lengths of comment blocks. Defensive strategy
211 * against the exploit:
212 * Microsoft Security Bulletin MS04-028
213 * Buffer Overrun in JPEG Processing (GDI+) Could
214 * Allow Code Execution (833987)
215 * Enabled with +inspect-jpegs in actions files.
217 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
218 * - Fixed usage info for non-unix platforms
219 * - Fixed small cmdline parsing bug
221 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
222 * Diagnostics: Now preserve the returncode of pthread_create
223 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
225 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
226 * No longer ignore SIGCHLD. Fixes bug #769381
228 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
229 * Close jarfile on graceful exit, remove stray line
231 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
232 * Cosmetics: Killed a warning, a typo and an allocation left at exit
234 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
235 * No longer rely on non-POSIX.1 extensions of getcwd().
238 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
239 * Replaced setenv() by posix-compliant putenv()
240 * Thanks to Neil McCalden (nmcc AT users.sf.net).
242 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
243 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
245 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
246 * Clean-up and extension of improvements for forked mode:
247 * - Child's return code now consists of flags RC_FLAG_*
248 * - Reporting toggle to parent now properly #ifdef'ed
249 * - Children now report blocking to parent. This enables
250 * statistics in forked mode
252 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
253 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
254 * Children now report having been toggled through _exit(17), parents
255 * watch for that code and toggle themselves if found.
257 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
258 * Wrapping all *_r functions (the non-_r versions of them) with
259 * mutex semaphores for OSX. Hopefully this will take care of all
260 * of those pesky crash reports.
262 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
263 * Fixed two mostly harmless mem leaks
265 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
266 * Fix: Head of global clients list now initialized to NULL
268 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
269 * Added basic support for OPTIONS and TRACE HTTP methods:
270 * - New interceptor direct_response() added in chat().
271 * - sed() moved to earlier in the process, so that the
272 * Host: header is evaluated before actions and forwarding
275 * Revision 1.92 2002/05/08 16:00:46 oes
276 * Chat's buffer handling:
277 * - Fixed bug with unchecked out-of-mem conditions
278 * while reading client request & server headers
279 * - No longer predict if the buffer limit will be exceeded
280 * in the next read -- check add_to_iob's new
281 * return code. If buffer couldn't be extended
282 * (policy or out-of-mem) while
283 * - reading from client: abort
284 * - reading server headers: send error page
285 * - buffering server body for filter: flush,
286 * and if that fails: send error page
288 * Revision 1.91 2002/04/08 20:35:58 swa
291 * Revision 1.90 2002/04/02 14:57:28 oes
292 * Made sending wafers independent of FEATURE_COOKIE_JAR
294 * Revision 1.89 2002/03/31 17:18:59 jongfoster
295 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
297 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
298 * More compiler warning message maintenance
300 * Revision 1.87 2002/03/26 22:29:54 swa
301 * we have a new homepage!
303 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
304 * Workaround for closing the jarfile before load_config() comes around again
306 * Revision 1.85 2002/03/24 15:23:33 jongfoster
309 * Revision 1.84 2002/03/24 13:25:43 swa
310 * name change related issues
312 * Revision 1.83 2002/03/16 23:54:06 jongfoster
313 * Adding graceful termination feature, to help look for memory leaks.
314 * If you enable this (which, by design, has to be done by hand
315 * editing config.h) and then go to http://i.j.b/die, then the program
316 * will exit cleanly after the *next* request. It should free all the
317 * memory that was used.
319 * Revision 1.82 2002/03/13 00:27:05 jongfoster
322 * Revision 1.81 2002/03/12 01:42:50 oes
323 * Introduced modular filters
325 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
326 * OS/2 port maintenance:
327 * - Fixed EMX build - it had decayed a little
328 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
329 * substituted a memset for now.
331 * Revision 1.79 2002/03/09 20:03:52 jongfoster
332 * - Making various functions return int rather than size_t.
333 * (Undoing a recent change). Since size_t is unsigned on
334 * Windows, functions like read_socket that return -1 on
335 * error cannot return a size_t.
337 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
338 * crashes, and also frequently caused JB to jump to 100%
339 * CPU and stay there. (Because it thought it had just
340 * read ((unsigned)-1) == 4Gb of data...)
342 * - The signature of write_socket has changed, it now simply
343 * returns success=0/failure=nonzero.
345 * - Trying to get rid of a few warnings --with-debug on
346 * Windows, I've introduced a new type "jb_socket". This is
347 * used for the socket file descriptors. On Windows, this
348 * is SOCKET (a typedef for unsigned). Everywhere else, it's
349 * an int. The error value can't be -1 any more, so it's
350 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
351 * Windows it maps to the #define INVALID_SOCKET.)
353 * - The signature of bind_port has changed.
355 * Revision 1.78 2002/03/08 21:35:04 oes
356 * Added optional group supplement to --user option. Will now use default group of user if no group given
358 * Revision 1.77 2002/03/07 03:52:06 oes
359 * - Fixed compiler warnings etc
360 * - Improved handling of failed DNS lookups
362 * Revision 1.76 2002/03/06 22:54:35 jongfoster
363 * Automated function-comment nitpicking.
365 * Revision 1.75 2002/03/06 10:02:19 oes
366 * Fixed stupid bug when --user was not given
368 * Revision 1.74 2002/03/06 00:49:31 jongfoster
369 * Fixing warning on Windows
370 * Making #ifdefs that refer to the same variable consistently
371 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
373 * Revision 1.73 2002/03/05 23:57:30 hal9
374 * Stray character 's' on line 1618 was breaking build.
376 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
377 * - Re-enable OS/2 building after new parms were added
378 * - Fix false out of memory report when resolving CGI templates when no IP
379 * address is available of failed attempt (a la no such domain)
381 * Revision 1.71 2002/03/05 18:13:56 oes
382 * Added --user option
384 * Revision 1.70 2002/03/05 04:52:42 oes
385 * Deleted non-errlog debugging code
387 * Revision 1.69 2002/03/04 23:50:00 jongfoster
388 * Splitting off bind_port() call into bind_port_helper(), with
391 * Revision 1.68 2002/03/04 20:17:32 oes
394 * Revision 1.67 2002/03/04 18:18:57 oes
395 * - Removed _DEBUG mode
396 * - Cleand up cmdline parsing
397 * - Introduced --no-daemon, --pidfile options
398 * - Cleaned up signal handling:
399 * - Terminate cleanly on INT, TERM and ABRT
400 * - Schedule logfile for re-opening on HUP
401 * - Ignore CHLD and PIPE
402 * - Leave the rest with their default handlers
403 * - Uniform handler registration
404 * - Added usage() function
405 * - Played styleguide police
407 * Revision 1.66 2002/03/03 15:06:55 oes
408 * Re-enabled automatic config reloading
410 * Revision 1.65 2002/03/03 14:49:11 oes
411 * Fixed CLF logging: Now uses client's original HTTP request
413 * Revision 1.64 2002/03/03 09:18:03 joergs
414 * Made jumbjuster work on AmigaOS again.
416 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
417 * Clean up a little CRLF unpleasantness that suddenly appeared
419 * Revision 1.62 2002/02/20 23:17:23 jongfoster
420 * Detecting some out-of memory conditions and exiting with a log message.
422 * Revision 1.61 2002/01/17 21:01:52 jongfoster
423 * Moving all our URL and URL pattern parsing code to urlmatch.c.
425 * Revision 1.60 2001/12/30 14:07:32 steudten
426 * - Add signal handling (unix)
427 * - Add SIGHUP handler (unix)
428 * - Add creation of pidfile (unix)
429 * - Add action 'top' in rc file (RH)
430 * - Add entry 'SIGNALS' to manpage
431 * - Add exit message to logfile (unix)
433 * Revision 1.59 2001/12/13 14:07:18 oes
434 * Fixed Bug: 503 error page now sent OK
436 * Revision 1.58 2001/11/30 23:37:24 jongfoster
437 * Renaming the Win32 config file to config.txt - this is almost the
438 * same as the corresponding UNIX name "config"
440 * Revision 1.57 2001/11/16 00:47:43 jongfoster
441 * Changing the tty-disconnection code to use setsid().
443 * Revision 1.56 2001/11/13 20:20:54 jongfoster
444 * Tabs->spaces, fixing a bug with missing {} around an if()
446 * Revision 1.55 2001/11/13 20:14:53 jongfoster
447 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
449 * Revision 1.54 2001/11/07 00:03:14 steudten
450 * Give reliable return value if an error
451 * occurs not just 0 with new daemon mode.
453 * Revision 1.53 2001/11/05 21:41:43 steudten
454 * Add changes to be a real daemon just for unix os.
455 * (change cwd to /, detach from controlling tty, set
456 * process group and session leader to the own process.
458 * Add some fatal-error log message for failed malloc().
459 * Add '-d' if compiled with 'configure --with-debug' to
460 * enable debug output.
462 * Revision 1.52 2001/10/26 20:11:20 jongfoster
463 * Fixing type mismatch
465 * Revision 1.51 2001/10/26 17:38:28 oes
468 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
469 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
470 * threads to call select() simultaneously. So, it's time to do a real, live,
471 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
472 * (native). Both versions will work, but using __OS2__ offers multi-threading.
474 * Revision 1.49 2001/10/23 21:41:35 jongfoster
475 * Added call to initialize the (statically-allocated of course)
476 * "out of memory" CGI response.
478 * Revision 1.48 2001/10/10 19:56:46 jongfoster
479 * Moving some code that wasn't cookie-related out of an #ifdef
482 * Revision 1.47 2001/10/10 16:44:36 oes
483 * Added CONNECT destination port limitation check
485 * Revision 1.46 2001/10/08 15:17:41 oes
486 * Re-enabled SSL forwarding
488 * Revision 1.45 2001/10/07 15:42:11 oes
489 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
491 * Moved downgrading of the HTTP version from parse_http_request to
492 * chat(), since we can't decide if it is necessary before we have
493 * determined the actions for the URL. The HTTP command is now
494 * *always* re-built so the repairs need no longer be special-cased.
496 * filter_popups now gets a csp pointer so it can raise the new
497 * CSP_FLAG_MODIFIED flag.
501 * Added configurable size limit for the IOB. If the IOB grows so
502 * large that the next read would exceed the limit, the header
503 * is generated, and the header & unfiltered buffer are flushed
504 * to the client. Chat then continues in non-buffering,
505 * non-filtering body mode.
507 * Revision 1.44 2001/10/02 18:13:57 oes
510 * Revision 1.43 2001/10/02 15:32:13 oes
511 * Moved generation of hdr
513 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
514 * Cleaning up 2 compiler warnings on OS/2.
516 * Revision 1.41 2001/09/16 17:05:14 jongfoster
517 * Removing unused #include showarg.h
519 * Revision 1.40 2001/09/16 15:41:45 jongfoster
520 * Fixing signed/unsigned comparison warning.
522 * Revision 1.39 2001/09/16 13:21:27 jongfoster
523 * Changes to use new list functions.
525 * Revision 1.38 2001/09/16 13:01:46 jongfoster
526 * Removing redundant function call that zeroed zalloc()'d memory.
528 * Revision 1.37 2001/09/10 11:12:24 oes
529 * Deleted unused variable
531 * Revision 1.36 2001/09/10 10:56:15 oes
532 * Silenced compiler warnings
534 * Revision 1.35 2001/07/31 14:44:22 oes
535 * Deleted unused size parameter from filter_popups()
537 * Revision 1.34 2001/07/30 22:08:36 jongfoster
538 * Tidying up #defines:
539 * - All feature #defines are now of the form FEATURE_xxx
540 * - Permanently turned off WIN_GUI_EDIT
541 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
543 * Revision 1.33 2001/07/29 19:32:00 jongfoster
544 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
546 * Revision 1.32 2001/07/29 18:47:05 jongfoster
547 * Adding missing #include "loadcfg.h"
549 * Revision 1.31 2001/07/29 12:17:48 oes
550 * Applied pthread fix by Paul Lieverse
552 * Revision 1.30 2001/07/25 22:57:13 jongfoster
553 * __BEOS__ no longer overrides FEATURE_PTHREAD.
554 * This is because FEATURE_PTHREAD will soon be widely used, so I
555 * want to keep it simple.
557 * Revision 1.29 2001/07/24 12:47:06 oes
558 * Applied BeOS support update by Eugenia
560 * Revision 1.28 2001/07/23 13:26:12 oes
561 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
563 * Revision 1.27 2001/07/19 19:09:47 haroon
564 * - Added code to take care of the situation where while processing the first
565 * server response (which includes the server header), after finding the end
566 * of the headers we were not looking past the end of the headers for
567 * content modification. I enabled it for filter_popups.
568 * Someone else should look to see if other similar operations should be
569 * done to the discarded portion of the buffer.
571 * Note 2001/07/20: No, the other content modification mechanisms will process
572 * the whole iob later anyway. --oes
574 * Revision 1.26 2001/07/18 12:31:36 oes
577 * Revision 1.25 2001/07/15 19:43:49 jongfoster
578 * Supports POSIX threads.
579 * Also removed some unused #includes.
581 * Revision 1.24 2001/07/13 14:00:40 oes
582 * - Generic content modification scheme:
583 * Each feature has its own applicability flag that is set
584 * from csp->action->flags.
585 * Replaced the "filtering" int flag , by a function pointer
586 * "content_filter" to the function that will do the content
587 * modification. If it is != NULL, the document will be buffered
588 * and processed through *content_filter, which must set
589 * csp->content_length and return a modified copy of the body
590 * or return NULL (on failiure).
591 * - Changed csp->is_text to the more generic bitmap csp->content_type
592 * which can currently take the valued CT_TEXT or CT_GIF
594 * - Removed all #ifdef PCRS
596 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
597 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
598 * the `block_acl' call. This prevents a compilation error when the user
599 * does not wish to use the "ACL" feature.
601 * Revision 1.22 2001/06/29 21:45:41 oes
602 * Indentation, CRLF->LF, Tab-> Space
604 * Revision 1.21 2001/06/29 13:29:36 oes
605 * - Cleaned up, improved comments
606 * - Unified all possible interceptors (CGI,
607 * block, trust, fast_redirect) in one
608 * place, with one (CGI) answer generation
609 * mechansim. Much clearer now.
610 * - Removed the GIF image generation, which
611 * is now done in filters.c:block_url()
612 * - Made error conditions like domain lookup
613 * failiure or (various) problems while talking
614 * to the server use cgi.c:error_response()
615 * instead of generating HTML/HTTP in chat() (yuck!)
616 * - Removed logentry from cancelled commit
618 * Revision 1.20 2001/06/09 10:55:28 jongfoster
619 * Changing BUFSIZ ==> BUFFER_SIZE
621 * Revision 1.19 2001/06/07 23:12:52 jongfoster
622 * Replacing function pointer in struct gateway with a directly
623 * called function forwarded_connect().
624 * Replacing struct gateway with struct forward_spec
626 * Revision 1.18 2001/06/03 19:12:16 oes
627 * introduced new cgi handling
629 * Revision 1.17 2001/06/01 20:07:23 jongfoster
630 * Now uses action +image-blocker{} rather than config->tinygif
632 * Revision 1.16 2001/06/01 18:49:17 jongfoster
633 * Replaced "list_share" with "list" - the tiny memory gain was not
634 * worth the extra complexity.
636 * Revision 1.15 2001/05/31 21:24:47 jongfoster
637 * Changed "permission" to "action" throughout.
638 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
639 * Moved vanilla wafer check into chat(), since we must now
640 * decide whether or not to add it based on the URL.
642 * Revision 1.14 2001/05/29 20:14:01 joergs
643 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
646 * Revision 1.13 2001/05/29 09:50:24 jongfoster
647 * Unified blocklist/imagelist/permissionslist.
648 * File format is still under discussion, but the internal changes
651 * Also modified interceptor behaviour:
652 * - We now intercept all URLs beginning with one of the following
653 * prefixes (and *only* these prefixes):
655 * * http://ijbswa.sf.net/config/
656 * * http://ijbswa.sourceforge.net/config/
657 * - New interceptors "home page" - go to http://i.j.b/ to see it.
658 * - Internal changes so that intercepted and fast redirect pages
659 * are not replaced with an image.
660 * - Interceptors now have the option to send a binary page direct
661 * to the client. (i.e. ijb-send-banner uses this)
662 * - Implemented show-url-info interceptor. (Which is why I needed
663 * the above interceptors changes - a typical URL is
664 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
665 * The previous mechanism would not have intercepted that, and
666 * if it had been intercepted then it then it would have replaced
669 * Revision 1.12 2001/05/27 22:17:04 oes
671 * - re_process_buffer no longer writes the modified buffer
672 * to the client, which was very ugly. It now returns the
673 * buffer, which it is then written by chat.
675 * - content_length now adjusts the Content-Length: header
676 * for modified documents rather than crunch()ing it.
677 * (Length info in csp->content_length, which is 0 for
678 * unmodified documents)
680 * - For this to work, sed() is called twice when filtering.
682 * Revision 1.11 2001/05/26 17:27:53 jongfoster
683 * Added support for CLF and fixed LOG_LEVEL_LOG.
684 * Also did CRLF->LF fix of my previous patch.
686 * Revision 1.10 2001/05/26 15:26:15 jongfoster
687 * ACL feature now provides more security by immediately dropping
688 * connections from untrusted hosts.
690 * Revision 1.9 2001/05/26 00:28:36 jongfoster
691 * Automatic reloading of config file.
692 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
693 * Most of the global variables have been moved to a new
694 * struct configuration_spec, accessed through csp->config->globalname
695 * Most of the globals remaining are used by the Win32 GUI.
697 * Revision 1.8 2001/05/25 22:43:18 jongfoster
698 * Fixing minor memory leak and buffer overflow.
700 * Revision 1.7 2001/05/25 22:34:30 jongfoster
703 * Revision 1.6 2001/05/23 00:13:58 joergs
704 * AmigaOS support fixed.
706 * Revision 1.5 2001/05/22 18:46:04 oes
708 * - Enabled filtering banners by size rather than URL
709 * by adding patterns that replace all standard banner
710 * sizes with the "Junkbuster" gif to the re_filterfile
712 * - Enabled filtering WebBugs by providing a pattern
713 * which kills all 1x1 images
715 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
716 * which is selected by the (nonstandard and therefore
717 * capital) letter 'U' in the option string.
718 * It causes the quantifiers to be ungreedy by default.
719 * Appending a ? turns back to greedy (!).
721 * - Added a new interceptor ijb-send-banner, which
722 * sends back the "Junkbuster" gif. Without imagelist or
723 * MSIE detection support, or if tinygif = 1, or the
724 * URL isn't recognized as an imageurl, a lame HTML
725 * explanation is sent instead.
727 * - Added new feature, which permits blocking remote
728 * script redirects and firing back a local redirect
730 * The feature is conditionally compiled, i.e. it
731 * can be disabled with --disable-fast-redirects,
732 * plus it must be activated by a "fast-redirects"
733 * line in the config file, has its own log level
734 * and of course wants to be displayed by show-proxy-args
735 * Note: Boy, all the #ifdefs in 1001 locations and
736 * all the fumbling with configure.in and acconfig.h
737 * were *way* more work than the feature itself :-(
739 * - Because a generic redirect template was needed for
740 * this, tinygif = 3 now uses the same.
742 * - Moved GIFs, and other static HTTP response templates
747 * - Removed some >400 CRs again (Jon, you really worked
750 * Revision 1.4 2001/05/21 19:34:01 jongfoster
751 * Made failure to bind() a fatal error.
753 * Revision 1.3 2001/05/20 01:21:20 jongfoster
754 * Version 2.9.4 checkin.
755 * - Merged popupfile and cookiefile, and added control over PCRS
756 * filtering, in new "permissionsfile".
757 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
758 * file error you now get a message box (in the Win32 GUI) rather
759 * than the program exiting with no explanation.
760 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
762 * - Removed tabs from "config"
763 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
764 * - Bumped up version number.
766 * Revision 1.2 2001/05/17 22:34:44 oes
767 * - Added hint on GIF char array generation to jcc.c
768 * - Cleaned CRLF's from the sources and related files
769 * - Repaired logging for REF and FRC
771 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
772 * Initial import of version 2.9.3 source tree
775 *********************************************************************/
781 #include <sys/types.h>
790 # ifndef FEATURE_PTHREAD
794 # include <windows.h>
795 # include <process.h>
796 # endif /* ndef FEATURE_PTHREAD */
799 # ifndef _WIN_CONSOLE
801 # endif /* ndef _WIN_CONSOLE */
802 # include "w32svrapi.h"
804 #else /* ifndef _WIN32 */
806 # if !defined (__OS2__)
808 # include <sys/wait.h>
809 # endif /* ndef __OS2__ */
810 # include <sys/time.h>
811 # include <sys/stat.h>
812 # include <sys/ioctl.h>
815 #include <sys/termios.h>
826 # include <socket.h> /* BeOS has select() for sockets only. */
827 # include <OS.h> /* declarations for threads and stuff. */
830 # if defined(__EMX__) || defined(__OS2__)
831 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
836 #define bzero(B,N) memset(B,0x00,n)
851 #include "killpopup.h"
852 #include "miscutil.h"
854 #include "jbsockets.h"
859 #include "urlmatch.h"
861 const char jcc_h_rcs[] = JCC_H_VERSION;
862 const char project_h_rcs[] = PROJECT_H_VERSION;
865 struct client_state clients[1];
866 struct file_list files[1];
868 #ifdef FEATURE_STATISTICS
869 int urls_read = 0; /* total nr of urls read inc rejected */
870 int urls_rejected = 0; /* total nr of urls rejected */
871 #endif /* def FEATURE_STATISTICS */
873 #ifdef FEATURE_GRACEFUL_TERMINATION
877 static void listen_loop(void);
878 static void chat(struct client_state *csp);
880 void serve(struct client_state *csp);
881 #else /* ifndef AMIGA */
882 static void serve(struct client_state *csp);
883 #endif /* def AMIGA */
886 static int32 server_thread(void *data);
887 #endif /* def __BEOS__ */
890 #define sleep(N) Sleep(((N) * 1000))
894 #define sleep(N) DosSleep(((N) * 100))
897 #ifdef FEATURE_PTHREAD
898 pthread_mutex_t log_mutex;
899 pthread_mutex_t log_init_mutex;
901 #ifndef HAVE_GMTIME_R
902 pthread_mutex_t gmtime_mutex;
903 #endif /* ndef HAVE_GMTIME_R */
905 #ifndef HAVE_LOCALTIME_R
906 pthread_mutex_t localtime_mutex;
907 #endif /* ndef HAVE_GMTIME_R */
909 #ifndef HAVE_GETHOSTBYADDR_R
910 pthread_mutex_t gethostbyaddr_mutex;
911 #endif /* ndef HAVE_GETHOSTBYADDR_R */
913 #ifndef HAVE_GETHOSTBYNAME_R
914 pthread_mutex_t gethostbyname_mutex;
915 #endif /* ndef HAVE_GETHOSTBYNAME_R */
918 pthread_mutex_t rand_mutex;
919 #endif /* ndef HAVE_RANDOM */
921 #endif /* FEATURE_PTHREAD */
923 #if defined(unix) || defined(__EMX__)
924 const char *basedir = NULL;
925 const char *pidfile = NULL;
926 int received_hup_signal = 0;
927 #endif /* defined unix */
929 /* The vanilla wafer. */
930 static const char VANILLA_WAFER[] =
931 "NOTICE=TO_WHOM_IT_MAY_CONCERN_"
932 "Do_not_send_me_any_copyrighted_information_other_than_the_"
933 "document_that_I_am_requesting_or_any_of_its_necessary_components._"
934 "In_particular_do_not_send_me_any_cookies_that_"
935 "are_subject_to_a_claim_of_copyright_by_anybody._"
936 "Take_notice_that_I_refuse_to_be_bound_by_any_license_condition_"
937 "(copyright_or_otherwise)_applying_to_any_cookie._";
939 /* HTTP snipplets. */
940 const char CSUCCEED[] =
941 "HTTP/1.0 200 Connection established\n"
942 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
944 const char CHEADER[] =
945 "HTTP/1.0 400 Invalid header received from browser\r\n"
946 "Connection: close\r\n\r\n"
947 "Invalid header received from browser.\r\n";
949 const char CFORBIDDEN[] =
950 "HTTP/1.0 403 Connection not allowable\r\n"
951 "Proxy-Agent: Privoxy " VERSION "\r\n"
952 "X-Hint: If you read this message interactively, then you know why this happens ,-)\r\n"
953 "Connection: close\r\n\r\n";
955 const char FTP_RESPONSE[] =
956 "HTTP/1.0 400 Invalid request received from browser\r\n"
957 "Connection: close\r\n\r\n"
958 "Invalid request. Privoxy doesn't support FTP.\r\n";
960 const char GOPHER_RESPONSE[] =
961 "HTTP/1.0 400 Invalid request received from browser\r\n"
962 "Connection: close\r\n\r\n"
963 "Invalid request. Privoxy doesn't support gopher.\r\n";
965 /* XXX: should be a template */
966 const char MISSING_DESTINATION_RESPONSE[] =
967 "HTTP/1.0 400 Bad request received from browser\r\n"
968 "Proxy-Agent: Privoxy " VERSION "\r\n"
969 "Connection: close\r\n\r\n"
970 "Bad request. Privoxy was unable to extract the destination.\r\n";
972 /* XXX: should be a template */
973 const char NO_SERVER_DATA_RESPONSE[] =
974 "HTTP/1.0 502 Server or forwarder response empty\r\n"
975 "Proxy-Agent: Privoxy " VERSION "\r\n"
976 "Connection: close\r\n\r\n"
977 "Empty server or forwarder response.\r\n"
978 "The connection was closed without sending any data.\r\n";
980 /* XXX: should be a template */
981 const char NULL_BYTE_RESPONSE[] =
982 "HTTP/1.0 400 Bad request received from browser\r\n"
983 "Proxy-Agent: Privoxy " VERSION "\r\n"
984 "Connection: close\r\n\r\n"
985 "Bad request. Null byte(s) before end of request.\r\n";
987 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
988 /*********************************************************************
990 * Function : sig_handler
992 * Description : Signal handler for different signals.
993 * Exit gracefully on TERM and INT
994 * or set a flag that will cause the errlog
995 * to be reopened by the main thread on HUP.
998 * 1 : the_signal = the signal cause this function to call
1002 *********************************************************************/
1003 static void sig_handler(int the_signal)
1009 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1020 received_hup_signal = 1;
1025 * We shouldn't be here, unless we catch signals
1026 * in main() that we can't handle here!
1028 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1036 /*********************************************************************
1038 * Function : client_protocol_is_unsupported
1040 * Description : Checks if the client used a known unsupported
1041 * protocol and deals with it by sending an error
1045 * 1 : csp = Current client state (buffers, headers, etc...)
1046 * 2 : req = the first request line send by the client
1048 * Returns : TRUE if an error response has been generated, or
1049 * FALSE if the request doesn't look invalid.
1051 *********************************************************************/
1052 int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1054 char buf[BUFFER_SIZE];
1057 * If it's a FTP or gopher request, we don't support it.
1059 * These checks are better than nothing, but they might
1060 * not work in all configurations and some clients might
1061 * have problems digesting the answer.
1063 * They should, however, never cause more problems than
1064 * Privoxy's old behaviour (returning the misleading HTML
1067 * "Could not resolve http://(ftp|gopher)://example.org").
1069 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1071 if (!strncmpic(req, "GET ftp://", 10))
1073 strcpy(buf, FTP_RESPONSE);
1074 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1075 csp->ip_addr_str, req);
1079 strcpy(buf, GOPHER_RESPONSE);
1080 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1081 csp->ip_addr_str, req);
1083 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1085 write_socket(csp->cfd, buf, strlen(buf));
1094 /*********************************************************************
1096 * Function : get_request_destination_elsewhere
1098 * Description : If the client's request was redirected into
1099 * Privoxy without the client's knowledge,
1100 * the request line lacks the destination host.
1102 * This function tries to get it elsewhere,
1103 * provided accept-intercepted-requests is enabled.
1105 * "Elsewhere" currently only means "Host: header",
1106 * but in the future we may ask the redirecting
1107 * packet filter to look the destination up.
1109 * If the destination stays unknown, an error
1110 * response is send to the client and headers
1111 * are freed so that chat() can return directly.
1114 * 1 : csp = Current client state (buffers, headers, etc...)
1115 * 2 : headers = a header list
1117 * Returns : JB_ERR_OK if the destination is now known, or
1118 * JB_ERR_PARSE if it isn't.
1120 *********************************************************************/
1121 jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1123 char buf[BUFFER_SIZE];
1126 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1128 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1129 " Privoxy isn't configured to accept intercepted requests.",
1130 csp->ip_addr_str, csp->http->cmd);
1131 /* XXX: Use correct size */
1132 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1133 csp->ip_addr_str, csp->http->cmd);
1135 strcpy(buf, CHEADER);
1136 write_socket(csp->cfd, buf, strlen(buf));
1137 destroy_list(headers);
1139 return JB_ERR_PARSE;
1141 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1143 /* Split the domain we just got for pattern matching */
1144 init_domain_components(csp->http);
1150 /* We can't work without destination. Go spread the news.*/
1152 req = list_to_text(headers);
1154 /* XXX: Use correct size */
1155 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1156 csp->ip_addr_str, csp->http->cmd);
1157 log_error(LOG_LEVEL_ERROR,
1158 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1159 csp->ip_addr_str, csp->http->cmd, req);
1162 strcpy(buf, MISSING_DESTINATION_RESPONSE);
1163 write_socket(csp->cfd, buf, strlen(buf));
1164 destroy_list(headers);
1166 return JB_ERR_PARSE;
1169 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1170 * to get the destination IP address, use it as host directly
1171 * or do a reverse DNS lookup first.
1176 /*********************************************************************
1178 * Function : get_server_headers
1180 * Description : Parses server headers in iob and fills them
1181 * into csp->headers so that they can later be
1185 * 1 : csp = Current client state (buffers, headers, etc...)
1187 * Returns : JB_ERR_OK if everything went fine, or
1188 * JB_ERR_PARSE if the headers were incomplete.
1190 *********************************************************************/
1191 jb_err get_server_headers(struct client_state *csp)
1193 int continue_hack_in_da_house = 0;
1196 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1201 * continue hack in da house. Ignore the ending of
1202 * this head and continue enlisting header lines.
1203 * The reason is described below.
1205 enlist(csp->headers, "");
1206 continue_hack_in_da_house = 0;
1209 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1212 * It's a bodyless continue response, don't
1213 * stop header parsing after reaching it's end.
1215 * As a result Privoxy will concatenate the
1216 * next response's head and parse and deliver
1217 * the headers as if they belonged to one request.
1219 * The client will separate them because of the
1220 * empty line between them.
1222 * XXX: What we're doing here is clearly against
1223 * the intended purpose of the continue header,
1224 * and under some conditions (HTTP/1.0 client request)
1225 * it's a standard violation.
1227 * Anyway, "sort of against the spec" is preferable
1228 * to "always getting confused by Continue responses"
1229 * (Privoxy's behaviour before this hack was added)
1231 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1232 continue_hack_in_da_house = 1;
1234 else if (*header == '\0')
1237 * If the header is empty, but the Continue hack
1238 * isn't active, we can assume that we reached the
1239 * end of the buffer before we hit the end of the
1242 * Inform the caller an let it decide how to handle it.
1244 return JB_ERR_PARSE;
1248 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1251 * XXX: Should we quit the request and return a
1252 * out of memory error page instead?
1254 log_error(LOG_LEVEL_ERROR,
1255 "Out of memory while enlisting server headers. %s lost.",
1265 /*********************************************************************
1267 * Function : crunch_reason
1269 * Description : Translates the crunch reason code into a string.
1272 * 1 : rsp = a http_response
1274 * Returns : A string with the crunch reason or an error description.
1276 *********************************************************************/
1277 const char *crunch_reason(const struct http_response *rsp)
1279 char * reason = NULL;
1281 assert(rsp != NULL);
1284 return "Internal error while searching for crunch reason";
1287 switch (rsp->reason)
1289 case RSP_REASON_UNSUPPORTED:
1290 reason = "Unsupported HTTP feature";
1292 case RSP_REASON_BLOCKED:
1295 case RSP_REASON_UNTRUSTED:
1296 reason = "Untrusted";
1298 case RSP_REASON_REDIRECTED:
1299 reason = "Redirected";
1301 case RSP_REASON_CGI_CALL:
1302 reason = "CGI Call";
1304 case RSP_REASON_NO_SUCH_DOMAIN:
1305 reason = "DNS failure";
1307 case RSP_REASON_FORWARDING_FAILED:
1308 reason = "Forwarding failed";
1310 case RSP_REASON_CONNECT_FAILED:
1311 reason = "Connection failure";
1313 case RSP_REASON_OUT_OF_MEMORY:
1314 reason = "Out of memory (may mask other reasons)";
1317 reason = "No reason recorded";
1325 /*********************************************************************
1327 * Function : send_crunch_response
1329 * Description : Delivers already prepared response for
1330 * intercepted requests, logs the interception
1331 * and frees the response.
1334 * 1 : csp = Current client state (buffers, headers, etc...)
1335 * 1 : rsp = Fully prepared response. Will be freed on exit.
1337 * Returns : Nothing.
1339 *********************************************************************/
1340 void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1342 const struct http_request *http = csp->http;
1343 char status_code[4];
1345 assert(rsp != NULL);
1346 assert(rsp->head != NULL);
1351 * Not supposed to happen. If it does
1352 * anyway, treat it as an unknown error.
1354 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1355 /* return code doesn't matter */
1360 /* If rsp is still NULL, we have serious internal problems. */
1361 log_error(LOG_LEVEL_FATAL,
1362 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1366 * Extract the status code from the actual head
1367 * that was send to the client. It is the only
1368 * way to get it right for all requests, including
1369 * the fixed ones for out-of-memory problems.
1371 * A head starts like this: 'HTTP/1.1 200...'
1375 status_code[0] = rsp->head[9];
1376 status_code[1] = rsp->head[10];
1377 status_code[2] = rsp->head[11];
1378 status_code[3] = '\0';
1380 /* Write the answer to the client */
1381 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1382 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1384 /* There is nothing we can do about it. */
1385 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1388 /* Log that the request was crunched and why. */
1389 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1390 http->hostport, http->path, crunch_reason(rsp));
1391 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1392 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1394 /* Clean up and return */
1395 if (cgi_error_memory() != rsp)
1397 free_http_response(rsp);
1403 /*********************************************************************
1405 * Function : request_contains_null_bytes
1407 * Description : Checks for NULL bytes in the request and sends
1408 * an error message to the client if any were found.
1411 * 1 : csp = Current client state (buffers, headers, etc...)
1412 * 2 : buf = Data from the client's request to check.
1413 * 3 : len = The data length.
1415 * Returns : TRUE if the request contained one or more NULL bytes, or
1418 *********************************************************************/
1419 int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1421 size_t c_len; /* Request lenght when treated as C string */
1423 c_len = strlen(buf);
1428 * Null byte(s) found. Log the request,
1429 * return an error response and hang up.
1431 size_t tmp_len = c_len;
1436 * Replace NULL byte(s) with '°' characters
1437 * so the request can be logged as string.
1438 * XXX: Is there a better replacement character?
1441 tmp_len += strlen(buf+tmp_len);
1442 } while (tmp_len < len);
1444 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1445 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1446 log_error(LOG_LEVEL_HEADER,
1447 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1449 strcpy(buf, NULL_BYTE_RESPONSE);
1450 write_socket(csp->cfd, buf, strlen(buf));
1452 /* XXX: Log correct size */
1453 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1462 /*********************************************************************
1466 * Description : Once a connection to the client has been accepted,
1467 * this function is called (via serve()) to handle the
1468 * main business of the communication. When this
1469 * function returns, the caller must close the client
1472 * FIXME: chat is nearly thousand lines long.
1476 * 1 : csp = Current client state (buffers, headers, etc...)
1478 * Returns : On success, the number of bytes written are returned (zero
1479 * indicates nothing was written). On error, -1 is returned,
1480 * and errno is set appropriately. If count is zero and the
1481 * file descriptor refers to a regular file, 0 will be
1482 * returned without causing any other effect. For a special
1483 * file, the results are not portable.
1485 *********************************************************************/
1486 static void chat(struct client_state *csp)
1489 * This next lines are a little ugly, but they simplifies the if statements
1490 * below. Basically if TOGGLE, then we want the if to test if the
1491 * CSP_FLAG_TOGGLED_ON flag ist set, else we don't. And if FEATURE_FORCE_LOAD,
1492 * then we want the if to test for CSP_FLAG_FORCED , else we don't
1494 #ifdef FEATURE_TOGGLE
1495 # define IS_TOGGLED_ON_AND (csp->flags & CSP_FLAG_TOGGLED_ON) &&
1496 #else /* ifndef FEATURE_TOGGLE */
1497 # define IS_TOGGLED_ON_AND
1498 #endif /* ndef FEATURE_TOGGLE */
1499 #ifdef FEATURE_FORCE_LOAD
1500 # define IS_NOT_FORCED_AND !(csp->flags & CSP_FLAG_FORCED) &&
1501 #else /* ifndef FEATURE_FORCE_LOAD */
1502 # define IS_NOT_FORCED_AND
1503 #endif /* def FEATURE_FORCE_LOAD */
1505 #define IS_ENABLED_AND IS_TOGGLED_ON_AND IS_NOT_FORCED_AND
1507 char buf[BUFFER_SIZE];
1515 int ms_iis5_hack = 0;
1516 size_t byte_count = 0;
1517 int forwarded_connect_retries = 0;
1518 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1519 const struct forward_spec * fwd;
1520 struct http_request *http;
1521 int len; /* for buffer sizes (and negative error codes) */
1522 #ifdef FEATURE_KILL_POPUPS
1523 int block_popups; /* bool, 1==will block popups */
1524 int block_popups_now = 0; /* bool, 1==currently blocking popups */
1525 #endif /* def FEATURE_KILL_POPUPS */
1527 int pcrs_filter; /* bool, 1==will filter through pcrs */
1528 int gif_deanimate; /* bool, 1==will deanimate gifs */
1529 int jpeg_inspect; /* bool, 1==will inspect jpegs */
1531 /* Function that does the content filtering for the current request */
1532 char *(*content_filter)() = NULL;
1534 /* Skeleton for HTTP response, if we should intercept the request */
1535 struct http_response *rsp;
1537 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1538 struct list header_list;
1539 struct list *headers = &header_list;
1543 memset(buf, 0, sizeof(buf));
1546 * Read the client's request. Note that since we're not using select() we
1547 * could get blocked here if a client connected, then didn't say anything!
1552 len = read_socket(csp->cfd, buf, sizeof(buf)-1);
1554 if (len <= 0) break; /* error! */
1556 if (request_contains_null_bytes(csp, buf, len))
1558 /* NULL bytes found and dealt with, just hang up. */
1563 * If there is no memory left for buffering the
1564 * request, there is nothing we can do but hang up
1566 if (add_to_iob(csp, buf, len))
1571 req = get_header(csp);
1575 break; /* no HTTP request! */
1580 continue; /* more to come! */
1583 /* Does the request line look invalid? */
1584 if (client_protocol_is_unsupported(csp, req))
1587 * Yes. The request has already been
1588 * answered with a error response, the buffers
1589 * were freed and we're done with chatting.
1594 #ifdef FEATURE_FORCE_LOAD
1595 /* If this request contains the FORCE_PREFIX,
1596 * better get rid of it now and set the force flag --oes
1599 if (strstr(req, FORCE_PREFIX))
1601 strclean(req, FORCE_PREFIX);
1602 log_error(LOG_LEVEL_FORCE, "Enforcing request \"%s\".\n", req);
1603 csp->flags |= CSP_FLAG_FORCED;
1606 #endif /* def FEATURE_FORCE_LOAD */
1608 switch( parse_http_request(req, http, csp) )
1611 log_error(LOG_LEVEL_ERROR, "Out of memory while parsing request.");
1614 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", req);
1622 if (http->cmd == NULL)
1624 strcpy(buf, CHEADER);
1625 write_socket(csp->cfd, buf, strlen(buf));
1626 /* XXX: Use correct size */
1627 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1628 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
1630 free_http_request(http);
1634 /* grab the rest of the client's headers */
1638 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
1640 len = read_socket(csp->cfd, buf, sizeof(buf));
1643 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
1648 * If there is no memory left for buffering the
1649 * request, there is nothing we can do but hang up
1651 if (add_to_iob(csp, buf, len))
1658 if (p == NULL) break;
1665 if (http->host == NULL)
1668 * If we still don't know the request destination,
1669 * the request is invalid or the client uses
1670 * Privoxy without it's knowledge.
1672 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
1675 * Our attempts to get the request destination
1676 * elsewhere failed or Privoxy is configured
1677 * to only accept proxy requests.
1679 * An error response has already been send
1680 * and we're done here.
1686 /* decide how to route the HTTP request */
1688 if ((fwd = forward_url(http, csp)) == NULL)
1690 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
1691 /* Never get here - LOG_LEVEL_FATAL causes program exit */
1694 /* build the http request to send to the server
1695 * we have to do one of the following:
1697 * create = use the original HTTP request to create a new
1698 * HTTP request that has either the path component
1699 * without the http://domainspec (w/path) or the
1700 * full orininal URL (w/url)
1701 * Note that the path and/or the HTTP version may
1702 * have been altered by now.
1704 * connect = Open a socket to the host:port of the server
1705 * and short-circuit server and client socket.
1707 * pass = Pass the request unchanged if forwarding a CONNECT
1708 * request to a parent proxy. Note that we'll be sending
1709 * the CFAIL message ourselves if connecting to the parent
1710 * fails, but we won't send a CSUCCEED message if it works,
1711 * since that would result in a double message (ours and the
1712 * parent's). After sending the request to the parent, we simply
1715 * here's the matrix:
1718 * +--------+--------+
1720 * 0 | create | connect|
1722 * Forwarding +--------+--------+
1724 * 1 | create | pass |
1726 * +--------+--------+
1731 * Determine the actions for this URL
1733 #ifdef FEATURE_TOGGLE
1734 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
1736 /* Most compatible set of actions (i.e. none) */
1737 init_current_action(csp->action);
1740 #endif /* ndef FEATURE_TOGGLE */
1742 url_actions(http, csp);
1747 * Check if a CONNECT request is allowable:
1748 * In the absence of a +limit-connect action, allow only port 443.
1749 * If there is an action, allow whatever matches the specificaton.
1753 if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443)
1754 || (csp->action->flags & ACTION_LIMIT_CONNECT
1755 && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) )
1757 if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS)
1760 * The response may confuse some clients,
1761 * but makes unblocking easier.
1763 log_error(LOG_LEVEL_ERROR, "Marking suspicious CONNECT request from %s for blocking.",
1765 csp->action->flags |= ACTION_BLOCK;
1770 strcpy(buf, CFORBIDDEN);
1771 write_socket(csp->cfd, buf, strlen(buf));
1772 log_error(LOG_LEVEL_CONNECT, "Denying suspicious CONNECT request from %s", csp->ip_addr_str);
1773 log_error(LOG_LEVEL_CLF, "%s - - [%T] \" \" 403 0", csp->ip_addr_str);
1781 * Downgrade http version from 1.1 to 1.0 if +downgrade
1784 if ( (http->ssl == 0)
1785 && (!strcmpic(http->ver, "HTTP/1.1"))
1786 && (csp->action->flags & ACTION_DOWNGRADE))
1789 http->ver = strdup("HTTP/1.0");
1791 if (http->ver == NULL)
1793 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1798 * Save a copy of the original request for logging
1800 http->ocmd = strdup(http->cmd);
1802 if (http->ocmd == NULL)
1804 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
1808 * (Re)build the HTTP request for non-SSL requests.
1809 * If forwarding, use the whole URL, else, use only the path.
1815 http->cmd = strdup(http->gpc);
1816 string_append(&http->cmd, " ");
1818 if (fwd->forward_host)
1820 string_append(&http->cmd, http->url);
1824 string_append(&http->cmd, http->path);
1826 string_append(&http->cmd, " ");
1827 string_append(&http->cmd, http->ver);
1829 if (http->cmd == NULL)
1831 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1833 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", http->cmd);
1835 enlist(csp->headers, http->cmd);
1837 /* Append the previously read headers */
1838 list_append_list_unique(csp->headers, headers);
1839 destroy_list(headers);
1842 * If the user has not supplied any wafers, and the user has not
1843 * told us to suppress the vanilla wafer, then send the vanilla wafer.
1845 if (list_is_empty(csp->action->multi[ACTION_MULTI_WAFER])
1846 && ((csp->action->flags & ACTION_VANILLA_WAFER) != 0))
1848 enlist(csp->action->multi[ACTION_MULTI_WAFER], VANILLA_WAFER);
1852 #ifdef FEATURE_KILL_POPUPS
1853 block_popups = ((csp->action->flags & ACTION_NO_POPUPS) != 0);
1854 #endif /* def FEATURE_KILL_POPUPS */
1856 pcrs_filter = (csp->rlist != NULL) && /* There are expressions to be used */
1857 (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER]));
1859 gif_deanimate = ((csp->action->flags & ACTION_DEANIMATE) != 0);
1861 jpeg_inspect = ((csp->action->flags & ACTION_JPEG_INSPECT) != 0);
1864 * We have a request. Now, check to see if we need to
1865 * intercept it, i.e. If ..
1869 /* We may not forward the request by rfc2616 sect 14.31 */
1870 (NULL != (rsp = direct_response(csp)))
1872 /* or we are enabled and... */
1873 || (IS_ENABLED_AND (
1875 /* ..the request was blocked */
1876 ( NULL != (rsp = block_url(csp)))
1878 /* ..or untrusted */
1879 #ifdef FEATURE_TRUST
1880 || ( NULL != (rsp = trust_url(csp)))
1881 #endif /* def FEATURE_TRUST */
1883 /* ..or a redirect kicked in */
1884 || ( NULL != (rsp = redirect_url(csp)))
1888 * .. or a CGI call was detected and answered.
1890 * This check comes last to give the user the power
1891 * to deny acces to some (or all) of the cgi pages.
1893 || (NULL != (rsp = dispatch_cgi(csp)))
1898 * Deliver, log and free the interception
1899 * response. Afterwards we're done here.
1901 send_crunch_response(csp, rsp);
1902 #ifdef FEATURE_STATISTICS
1903 /* Count as a rejected request */
1904 csp->flags |= CSP_FLAG_REJECTED;
1905 #endif /* def FEATURE_STATISTICS */
1910 hdr = sed(client_patterns, add_client_headers, csp);
1913 /* FIXME Should handle error properly */
1914 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
1917 list_remove_all(csp->headers);
1919 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
1921 if (fwd->forward_host)
1923 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
1924 fwd->forward_host, fwd->forward_port, http->hostport);
1928 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
1931 /* here we connect to the server, gateway, or the forwarder */
1933 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
1934 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
1936 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
1937 forwarded_connect_retries, http->hostport);
1940 if (csp->sfd == JB_INVALID_SOCKET)
1942 if (fwd->type != SOCKS_NONE)
1945 rsp = error_response(csp, "forwarding-failed", errno);
1947 else if (errno == EINVAL)
1949 rsp = error_response(csp, "no-such-domain", errno);
1953 rsp = error_response(csp, "connect-failed", errno);
1954 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
1959 /* Write the answer to the client */
1962 send_crunch_response(csp, rsp);
1969 if (fwd->forward_host || (http->ssl == 0))
1971 /* write the client's (modified) header to the server
1972 * (along with anything else that may be in the buffer)
1975 if (write_socket(csp->sfd, hdr, strlen(hdr))
1976 || (flush_socket(csp->sfd, csp) < 0))
1978 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
1981 rsp = error_response(csp, "connect-failed", errno);
1985 send_crunch_response(csp, rsp);
1995 * We're running an SSL tunnel and we're not forwarding,
1996 * so just send the "connect succeeded" message to the
1997 * client, flush the rest, and get out of the way.
1999 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 0",
2000 csp->ip_addr_str, http->ocmd);
2002 if (write_socket(csp->cfd, CSUCCEED, sizeof(CSUCCEED)-1))
2010 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2012 /* we're finished with the client's header */
2015 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2017 /* pass data between the client and server
2018 * until one or the other shuts down the connection.
2027 * FD_ZERO here seems to point to an errant macro which crashes.
2028 * So do this by hand for now...
2030 memset(&rfds,0x00,sizeof(fd_set));
2034 FD_SET(csp->cfd, &rfds);
2035 FD_SET(csp->sfd, &rfds);
2037 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2041 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2045 /* this is the body of the browser's request
2046 * just read it and write it.
2049 if (FD_ISSET(csp->cfd, &rfds))
2051 len = read_socket(csp->cfd, buf, sizeof(buf));
2055 break; /* "game over, man" */
2058 if (write_socket(csp->sfd, buf, (size_t)len))
2060 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2067 * The server wants to talk. It could be the header or the body.
2068 * If `hdr' is null, then it's the header otherwise it's the body.
2069 * FIXME: Does `hdr' really mean `host'? No.
2073 if (FD_ISSET(csp->sfd, &rfds))
2076 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2080 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2082 if (http->ssl && (fwd->forward_host == NULL))
2085 * Just hang up. We already confirmed the client's CONNECT
2086 * request with status code 200 and unencrypted content is
2087 * no longer welcome.
2089 log_error(LOG_LEVEL_ERROR,
2090 "CONNECT already confirmed. Unable to tell the client about the problem.");
2094 rsp = error_response(csp, "connect-failed", errno);
2098 send_crunch_response(csp, rsp);
2104 /* Add a trailing zero. This lets filter_popups
2105 * use string operations.
2109 #ifdef FEATURE_KILL_POPUPS
2110 /* Filter the popups on this read. */
2111 if (block_popups_now)
2113 filter_popups(buf, csp);
2115 #endif /* def FEATURE_KILL_POPUPS */
2117 /* Normally, this would indicate that we've read
2118 * as much as the server has sent us and we can
2119 * close the client connection. However, Microsoft
2120 * in its wisdom has released IIS/5 with a bug that
2121 * prevents it from sending the trailing \r\n in
2122 * a 302 redirect header (and possibly other headers).
2123 * To work around this if we've haven't parsed
2124 * a full header we'll append a trailing \r\n
2125 * and see if this now generates a valid one.
2127 * This hack shouldn't have any impacts. If we've
2128 * already transmitted the header or if this is a
2129 * SSL connection, then we won't bother with this
2130 * hack. So we only work on partially received
2131 * headers. If we append a \r\n and this still
2132 * doesn't generate a valid header, then we won't
2133 * transmit anything to the client.
2138 if (server_body || http->ssl)
2141 * If we have been buffering up the document,
2142 * now is the time to apply content modification
2143 * and send the result to the client.
2148 * If the content filter fails, use the original
2149 * buffer and length.
2150 * (see p != NULL ? p : csp->iob->cur below)
2152 if (NULL == (p = (*content_filter)(csp)))
2154 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2157 hdr = sed(server_patterns_light, NULL, csp);
2161 /* FIXME Should handle error properly */
2162 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2165 if (write_socket(csp->cfd, hdr, strlen(hdr))
2166 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2168 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2178 break; /* "game over, man" */
2182 * This is NOT the body, so
2183 * Let's pretend the server just sent us a blank line.
2185 snprintf(buf, sizeof(buf), "\r\n");
2186 len = (int)strlen(buf);
2189 * Now, let the normal header parsing algorithm below do its
2190 * job. If it fails, we'll exit instead of continuing.
2197 * If this is an SSL connection or we're in the body
2198 * of the server document, just write it to the client,
2199 * unless we need to buffer the body for later content-filtering
2202 if (server_body || http->ssl)
2207 * If there is no memory left for buffering the content, or the buffer limit
2208 * has been reached, switch to non-filtering mode, i.e. make & write the
2209 * header, flush the iob and buf, and get out of the way.
2211 if (add_to_iob(csp, buf, len))
2216 log_error(LOG_LEVEL_ERROR, "Flushing header and buffers. Stepping back from filtering.");
2218 hdr = sed(server_patterns, add_server_headers, csp);
2222 * Memory is too tight to even generate the header.
2223 * Send our static "Out-of-memory" page.
2225 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2226 rsp = cgi_error_memory();
2227 send_crunch_response(csp, rsp);
2232 hdrlen = strlen(hdr);
2234 if (write_socket(csp->cfd, hdr, hdrlen)
2235 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
2236 || (write_socket(csp->cfd, buf, (size_t)len)))
2238 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2244 byte_count += hdrlen + (size_t)flushed + (size_t)len;
2246 content_filter = NULL;
2253 if (write_socket(csp->cfd, buf, (size_t)len))
2255 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2259 byte_count += (size_t)len;
2264 /* we're still looking for the end of the
2265 * server's header ... (does that make header
2266 * parsing an "out of body experience" ?
2270 * buffer up the data we just read. If that fails,
2271 * there's little we can do but send our static
2272 * out-of-memory page.
2274 if (add_to_iob(csp, buf, len))
2276 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2277 rsp = cgi_error_memory();
2278 send_crunch_response(csp, rsp);
2283 /* Convert iob into something sed() can digest */
2284 if (JB_ERR_PARSE == get_server_headers(csp))
2288 /* Well, we tried our MS IIS/5
2289 * hack and it didn't work.
2290 * The header is incomplete
2291 * and there isn't anything
2292 * we can do about it.
2298 /* Since we have to wait for
2299 * more from the server before
2300 * we can parse the headers
2301 * we just continue here.
2307 /* Did we actually get anything? */
2308 if (NULL == csp->headers->first)
2310 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2311 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2312 strcpy(buf, NO_SERVER_DATA_RESPONSE);
2313 write_socket(csp->cfd, buf, strlen(buf));
2314 free_http_request(http);
2318 /* we have now received the entire header.
2319 * filter it and send the result to the client
2322 hdr = sed(server_patterns, add_server_headers, csp);
2325 /* FIXME Should handle error properly */
2326 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2329 #ifdef FEATURE_KILL_POPUPS
2330 /* Start blocking popups if appropriate. */
2332 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2333 !http->ssl && /* We talk plaintext */
2334 block_popups) /* Policy allows */
2336 block_popups_now = 1;
2338 * Filter the part of the body that came in the same read
2339 * as the last headers:
2341 filter_popups(csp->iob->cur, csp);
2344 #endif /* def FEATURE_KILL_POPUPS */
2346 /* Buffer and pcrs filter this if appropriate. */
2348 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2349 !http->ssl && /* We talk plaintext */
2350 pcrs_filter) /* Policy allows */
2352 content_filter = pcrs_filter_response;
2355 /* Buffer and gif_deanimate this if appropriate. */
2357 if ((csp->content_type & CT_GIF) && /* It's an image/gif MIME-Type */
2358 !http->ssl && /* We talk plaintext */
2359 gif_deanimate) /* Policy allows */
2361 content_filter = gif_deanimate_response;
2364 /* Buffer and jpg_inspect this if appropriate. */
2366 if ((csp->content_type & CT_JPEG) && /* It's an image/jpeg MIME-Type */
2367 !http->ssl && /* We talk plaintext */
2368 jpeg_inspect) /* Policy allows */
2370 content_filter = jpeg_inspect_response;
2374 * Only write if we're not buffering for content modification
2376 if (!content_filter)
2378 /* write the server's (modified) header to
2379 * the client (along with anything else that
2380 * may be in the buffer)
2383 if (write_socket(csp->cfd, hdr, strlen(hdr))
2384 || ((len = flush_socket(csp->cfd, csp)) < 0))
2386 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2388 /* the write failed, so don't bother
2389 * mentioning it to the client...
2390 * it probably can't hear us anyway.
2396 byte_count += (size_t)len;
2399 /* we're finished with the server's header */
2404 /* If this was a MS IIS/5 hack then it means
2405 * the server has already closed the
2406 * connection. Nothing more to read. Time
2417 return; /* huh? we should never get here */
2420 if (csp->content_length == 0)
2423 * If Privoxy didn't recalculate the
2424 * Content-Lenght, byte_count is still
2427 csp->content_length = byte_count;
2430 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2431 csp->ip_addr_str, http->ocmd, csp->content_length);
2435 /*********************************************************************
2439 * Description : This is little more than chat. We only "serve" to
2440 * to close any socket that chat may have opened.
2443 * 1 : csp = Current client state (buffers, headers, etc...)
2447 *********************************************************************/
2449 void serve(struct client_state *csp)
2450 #else /* ifndef AMIGA */
2451 static void serve(struct client_state *csp)
2452 #endif /* def AMIGA */
2455 close_socket(csp->cfd);
2457 if (csp->sfd != JB_INVALID_SOCKET)
2459 close_socket(csp->sfd);
2462 csp->flags &= ~CSP_FLAG_ACTIVE;
2468 /*********************************************************************
2470 * Function : server_thread
2472 * Description : We only exist to call `serve' in a threaded environment.
2475 * 1 : data = Current client state (buffers, headers, etc...)
2477 * Returns : Always 0.
2479 *********************************************************************/
2480 static int32 server_thread(void *data)
2482 serve((struct client_state *) data);
2489 /*********************************************************************
2493 * Description : Print usage info & exit.
2495 * Parameters : Pointer to argv[0] for identifying ourselves
2499 *********************************************************************/
2500 void usage(const char *myname)
2502 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2504 "Usage: %s [--help] [--version] [configfile]\n"
2506 "Usage: %s [--help] [--version] [--no-daemon] [--pidfile pidfile] [--user user[.group]] [configfile]\n"
2508 "Aborting.\n", myname);
2515 /*********************************************************************
2517 * Function : initialize_mutexes
2519 * Description : Prepares mutexes if mutex support is available.
2523 * Returns : Void, exits in case of errors.
2525 *********************************************************************/
2526 void initialize_mutexes()
2530 #ifdef FEATURE_PTHREAD
2532 * Prepare global mutex semaphores
2534 err = pthread_mutex_init(&log_mutex, 0);
2536 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2539 * XXX: The assumptions below are a bit naive
2540 * and can cause locks that aren't necessary.
2542 * For example older FreeBSD versions (< 6.x?)
2543 * have no gethostbyname_r, but gethostbyname is
2546 #ifndef HAVE_GMTIME_R
2547 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2548 #endif /* ndef HAVE_GMTIME_R */
2550 #ifndef HAVE_LOCALTIME_R
2551 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2552 #endif /* ndef HAVE_GMTIME_R */
2554 #ifndef HAVE_GETHOSTBYADDR_R
2555 if (!err) err = pthread_mutex_init(&gethostbyaddr_mutex, 0);
2556 #endif /* ndef HAVE_GETHOSTBYADDR_R */
2558 #ifndef HAVE_GETHOSTBYNAME_R
2559 if (!err) err = pthread_mutex_init(&gethostbyname_mutex, 0);
2560 #endif /* ndef HAVE_GETHOSTBYNAME_R */
2563 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2564 #endif /* ndef HAVE_RANDOM */
2565 #endif /* FEATURE_PTHREAD */
2568 * TODO: mutex support for mingw32 would be swell.
2573 printf("Fatal error. Mutex initialization failed: %s.\n",
2582 /*********************************************************************
2586 * Description : Load the config file and start the listen loop.
2587 * This function is a lot more *sane* with the `load_config'
2588 * and `listen_loop' functions; although it stills does
2589 * a *little* too much for my taste.
2592 * 1 : argc = Number of parameters (including $0).
2593 * 2 : argv = Array of (char *)'s to the parameters.
2595 * Returns : 1 if : can't open config file, unrecognized directive,
2596 * stats requested in multi-thread mode, can't open the
2597 * log file, can't open the jar file, listen port is invalid,
2598 * any load fails, and can't bind port.
2600 * Else main never returns, the process must be signaled
2601 * to terminate execution. Or, on Windows, use the
2602 * "File", "Exit" menu option.
2604 *********************************************************************/
2606 int real_main(int argc, const char *argv[])
2608 int main(int argc, const char *argv[])
2612 unsigned int random_seed;
2614 struct passwd *pw = NULL;
2615 struct group *grp = NULL;
2624 #if !defined(_WIN32)
2632 * Parse the command line arguments
2634 while (++argc_pos < argc)
2637 /* Check to see if the service must be installed or uninstalled */
2638 if (strncmp(argv[argc_pos], "--install", 9) == 0)
2640 const char *pName = argv[argc_pos] + 9;
2643 exit( (install_service(pName)) ? 0 : 1 );
2645 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
2647 const char *pName = argv[argc_pos] + 11;
2650 exit((uninstall_service(pName)) ? 0 : 1);
2652 else if (strcmp(argv[argc_pos], "--service" ) == 0)
2654 bRunAsService = TRUE;
2655 w32_set_service_cwd();
2656 atexit(w32_service_exit_notify);
2659 #endif /* defined(_WIN32) */
2662 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2664 if (strcmp(argv[argc_pos], "--help") == 0)
2669 else if(strcmp(argv[argc_pos], "--version") == 0)
2671 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
2677 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
2682 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
2684 if (++argc_pos == argc) usage(argv[0]);
2685 pidfile = strdup(argv[argc_pos]);
2688 else if (strcmp(argv[argc_pos], "--user" ) == 0)
2690 if (++argc_pos == argc) usage(argv[argc_pos]);
2692 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
2695 if (NULL == (grp = getgrnam(p)))
2697 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
2701 if (NULL == (pw = getpwnam(argv[argc_pos])))
2703 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
2706 if (p != NULL) *--p = '\0';
2709 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
2714 #endif /* defined(unix) */
2715 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
2717 configfile = argv[argc_pos];
2720 } /* -END- while (more arguments) */
2723 if ( *configfile != '/' )
2725 char *abs_file, cwd[1024];
2727 /* make config-filename absolute here */
2728 if ( !(getcwd(cwd, sizeof(cwd))))
2730 perror("get working dir failed");
2734 if (!(basedir = strdup(cwd))
2735 || (!(abs_file = malloc( strlen( basedir ) + strlen( configfile ) + 5 ))))
2737 perror("malloc failed");
2740 strcpy( abs_file, basedir );
2741 strcat( abs_file, "/" );
2742 strcat( abs_file, configfile );
2743 configfile = abs_file;
2745 #endif /* defined unix */
2749 clients->next = NULL;
2753 #elif defined(_WIN32)
2757 /* Prepare mutexes if supported and necessary. */
2758 initialize_mutexes();
2760 random_seed = (unsigned int)time(NULL);
2762 srandom(random_seed);
2765 #endif /* ifdef HAVE_RANDOM */
2768 * Unix signal handling
2770 * Catch the abort, interrupt and terminate signals for a graceful exit
2771 * Catch the hangup signal so the errlog can be reopened.
2772 * Ignore the broken pipe signals (FIXME: Why?)
2774 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
2777 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
2778 const int ignored_signals[] = { SIGPIPE, 0 };
2780 for (idx = 0; catched_signals[idx] != 0; idx++)
2782 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
2783 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
2785 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
2786 #endif /* ifdef sun */
2788 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
2792 for (idx = 0; ignored_signals[idx] != 0; idx++)
2794 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
2796 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
2801 #else /* ifdef _WIN32 */
2802 # ifdef _WIN_CONSOLE
2804 * We *are* in a windows console app.
2805 * Print a verbose messages about FAQ's and such
2807 printf("%s", win32_blurb);
2808 # endif /* def _WIN_CONSOLE */
2809 #endif /* def _WIN32 */
2812 /* Initialize the CGI subsystem */
2813 cgi_init_error_messages();
2816 * If runnig on unix and without the --nodaemon
2817 * option, become a daemon. I.e. fork, detach
2818 * from tty and get process group leadership
2831 if ( pid < 0 ) /* error */
2836 else if ( pid != 0 ) /* parent */
2841 * must check for errors
2842 * child died due to missing files aso
2845 wpid = waitpid( pid, &status, WNOHANG );
2854 /* Should be more portable, but not as well tested */
2859 #else /* ndef __FreeBSD__ */
2861 #endif /* ndef __FreeBSD__ */
2862 fd = open("/dev/tty", O_RDONLY);
2865 /* no error check here */
2866 ioctl( fd, TIOCNOTTY,0 );
2870 /* FIXME: should close stderr (fd 2) here too, but the test
2872 * and load config file is done in listen_loop() and puts
2873 * some messages on stderr there.
2880 } /* -END- if (!no_daemon) */
2883 * As soon as we have written the PID file, we can switch
2884 * to the user and group ID indicated by the --user option
2890 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
2892 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
2898 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
2900 if (chroot(pw->pw_dir) < 0)
2902 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
2906 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
2909 if (setuid(pw->pw_uid))
2911 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
2915 char putenv_dummy[64];
2917 strcpy(putenv_dummy, "HOME=/");
2918 if (putenv(putenv_dummy) != 0)
2920 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
2923 snprintf(putenv_dummy, 64, "USER=%s", pw->pw_name);
2924 if (putenv(putenv_dummy) != 0)
2926 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
2932 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
2935 #endif /* defined unix */
2938 /* This will be FALSE unless the command line specified --service
2942 /* Yup, so now we must attempt to establish a connection
2943 * with the service dispatcher. This will only work if this
2944 * process was launched by the service control manager to
2945 * actually run as a service. If this isn't the case, i've
2946 * known it take around 30 seconds or so for the call to return.
2949 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
2950 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
2952 /* Service has run, and at this point is now being stopped, so just return */
2957 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
2959 /* An error occurred. Usually it's because --service was wrongly specified
2960 * and we were unable to connect to the Service Control Dispatcher because
2961 * it wasn't expecting us and is therefore not listening.
2963 * For now, just continue below to call the listen_loop function.
2966 #endif /* def _WIN32 */
2976 /*********************************************************************
2978 * Function : bind_port_helper
2980 * Description : Bind the listen port. Handles logging, and aborts
2984 * 1 : config = Privoxy configuration. Specifies port
2987 * Returns : Port that was opened.
2989 *********************************************************************/
2990 static jb_socket bind_port_helper(struct configuration_spec * config)
2995 if ( (config->haddr != NULL)
2996 && (config->haddr[0] == '1')
2997 && (config->haddr[1] == '2')
2998 && (config->haddr[2] == '7')
2999 && (config->haddr[3] == '.') )
3001 log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only",
3004 else if (config->haddr == NULL)
3006 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3011 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3012 config->hport, config->haddr);
3015 result = bind_port(config->haddr, config->hport, &bfd);
3022 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3023 "There may be another Privoxy or some other "
3024 "proxy running on port %d",
3025 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3026 config->hport, config->hport);
3029 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3030 "The hostname is not resolvable",
3031 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3034 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3035 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3038 /* shouldn't get here */
3039 return JB_INVALID_SOCKET;
3042 config->need_bind = 0;
3049 /* Without this simple workaround we get this compiler warning from _beginthread
3050 * warning C4028: formal parameter 1 different from declaration
3052 void w32_service_listen_loop(void *p)
3056 #endif /* def _WIN32 */
3059 /*********************************************************************
3061 * Function : listen_loop
3063 * Description : bind the listen port and enter a "FOREVER" listening loop.
3069 *********************************************************************/
3070 static void listen_loop(void)
3072 struct client_state *csp = NULL;
3074 struct configuration_spec * config;
3076 config = load_config();
3078 bfd = bind_port_helper(config);
3080 #ifdef FEATURE_GRACEFUL_TERMINATION
3081 while (!g_terminate)
3086 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3087 while (waitpid(-1, NULL, WNOHANG) > 0)
3089 /* zombie children */
3091 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3094 * Free data that was used by died threads
3100 * Re-open the errlog after HUP signal
3102 if (received_hup_signal)
3104 init_error_log(Argv[0], config->logfile, config->debug);
3105 received_hup_signal = 0;
3110 #ifdef FEATURE_COOKIE_JAR
3112 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3113 * already open. I think unload_configfile() is not being run, which should do
3114 * this work. Until that can get resolved, we'll use this workaround.
3118 if (csp->config->jar)
3120 fclose(csp->config->jar);
3121 csp->config->jar = NULL;
3123 #endif /* FEATURE_COOKIE_JAR */
3124 #endif /* __OS2__ */
3126 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3128 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3132 csp->flags |= CSP_FLAG_ACTIVE;
3133 csp->sfd = JB_INVALID_SOCKET;
3135 csp->config = config = load_config();
3137 if ( config->need_bind )
3140 * Since we were listening to the "old port", we will not see
3141 * a "listen" param change until the next IJB request. So, at
3142 * least 1 more request must be made for us to find the new
3143 * setting. I am simply closing the old socket and binding the
3146 * Which-ever is correct, we will serve 1 more page via the
3147 * old settings. This should probably be a "show-proxy-args"
3148 * request. This should not be a so common of an operation
3149 * that this will hurt people's feelings.
3154 bfd = bind_port_helper(config);
3157 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3159 if (!accept_connection(csp, bfd))
3161 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3174 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3177 #ifdef FEATURE_TOGGLE
3178 if (global_toggle_state)
3180 csp->flags |= CSP_FLAG_TOGGLED_ON;
3182 #endif /* def FEATURE_TOGGLE */
3184 if (run_loader(csp))
3186 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3187 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3191 if (block_acl(NULL,csp))
3193 log_error(LOG_LEVEL_CONNECT, "Connection dropped due to ACL");
3194 close_socket(csp->cfd);
3198 #endif /* def FEATURE_ACL */
3200 /* add it to the list of clients */
3201 csp->next = clients->next;
3202 clients->next = csp;
3204 if (config->multi_threaded)
3208 /* this is a switch () statment in the C preprocessor - ugh */
3209 #undef SELECTED_ONE_OPTION
3211 /* Use Pthreads in preference to native code */
3212 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3213 #define SELECTED_ONE_OPTION
3215 pthread_t the_thread;
3216 pthread_attr_t attrs;
3218 pthread_attr_init(&attrs);
3219 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3220 errno = pthread_create(&the_thread, &attrs,
3221 (void * (*)(void *))serve, csp);
3222 child_id = errno ? -1 : 0;
3223 pthread_attr_destroy(&attrs);
3227 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3228 #define SELECTED_ONE_OPTION
3229 child_id = _beginthread(
3230 (void (*)(void *))serve,
3235 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3236 #define SELECTED_ONE_OPTION
3237 child_id = _beginthread(
3238 (void(* _Optlink)(void*))serve,
3244 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3245 #define SELECTED_ONE_OPTION
3247 thread_id tid = spawn_thread
3248 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3250 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3252 child_id = (int) tid;
3261 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3262 #define SELECTED_ONE_OPTION
3263 csp->cfd = ReleaseSocket(csp->cfd, -1);
3266 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3267 NP_Output, Output(),
3268 NP_CloseOutput, FALSE,
3269 NP_Name, (ULONG)"privoxy child",
3273 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3274 NP_Output, Output(),
3275 NP_CloseOutput, FALSE,
3276 NP_Name, (ULONG)"privoxy child",
3277 NP_StackSize, 200*1024,
3283 ((struct Task *)child_id)->tc_UserData = csp;
3284 Signal((struct Task *)child_id, SIGF_SINGLE);
3289 #if !defined(SELECTED_ONE_OPTION)
3292 /* This block is only needed when using fork().
3293 * When using threads, the server thread was
3294 * created and run by the call to _beginthread().
3296 if (child_id == 0) /* child */
3299 #ifdef FEATURE_TOGGLE
3300 int inherited_toggle_state = global_toggle_state;
3301 #endif /* def FEATURE_TOGGLE */
3306 * If we've been toggled or we've blocked the request, tell Mom
3309 #ifdef FEATURE_TOGGLE
3310 if (inherited_toggle_state != global_toggle_state)
3312 rc |= RC_FLAG_TOGGLED;
3314 #endif /* def FEATURE_TOGGLE */
3316 #ifdef FEATURE_STATISTICS
3317 if (csp->flags & CSP_FLAG_REJECTED)
3319 rc |= RC_FLAG_BLOCKED;
3321 #endif /* ndef FEATURE_STATISTICS */
3325 else if (child_id > 0) /* parent */
3327 /* in a fork()'d environment, the parent's
3328 * copy of the client socket and the CSP
3332 #if !defined(_WIN32) && !defined(__CYGWIN__)
3334 wait( &child_status );
3337 * Evaluate child's return code: If the child has
3338 * - been toggled, toggle ourselves
3339 * - blocked its request, bump up the stats counter
3342 #ifdef FEATURE_TOGGLE
3343 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3345 global_toggle_state = !global_toggle_state;
3347 #endif /* def FEATURE_TOGGLE */
3349 #ifdef FEATURE_STATISTICS
3351 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3355 #endif /* def FEATURE_STATISTICS */
3357 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3358 close_socket(csp->cfd);
3359 csp->flags &= ~CSP_FLAG_ACTIVE;
3363 #undef SELECTED_ONE_OPTION
3364 /* end of cpp switch () */
3366 if (child_id < 0) /* failed */
3368 char buf[BUFFER_SIZE];
3370 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3372 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3374 write_socket(csp->cfd, buf, strlen(buf));
3375 close_socket(csp->cfd);
3376 csp->flags &= ~CSP_FLAG_ACTIVE;
3387 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3389 /* Clean up. Aim: free all memory (no leaks) */
3390 #ifdef FEATURE_GRACEFUL_TERMINATION
3392 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3394 unload_current_config_file();
3395 unload_current_actions_file();
3396 unload_current_re_filterfile();
3397 #ifdef FEATURE_TRUST
3398 unload_current_trust_file();
3401 if (config->multi_threaded)
3408 } while ((clients->next != NULL) && (--i > 0));
3412 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3423 #ifdef FEATURE_COOKIE_JAR
3424 if (NULL != config->jar)
3426 fclose(config->jar);
3430 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3431 /* Cleanup - remove taskbar icon etc. */
3436 #endif /* FEATURE_GRACEFUL_TERMINATION */