1 const char jcc_rcs[] = "$Id: jcc.c,v 1.125 2007/03/09 14:12:00 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 1.125 2007/03/09 14:12:00 fabiankeil
37 * - Move null byte check into separate function.
38 * - Don't confuse the client with error pages
39 * if a CONNECT request was already confirmed.
41 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
42 * Speed up NULL byte escaping and only log the complete
43 * NULL byte requests with header debugging enabled.
45 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
46 * Answer requests that contain NULL bytes with
47 * a custom response instead of waiting for more
48 * data until the client eventually hangs up.
50 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
51 * - Move delivery and logging of crunched responses
52 * from chat() into send_crunch_response().
53 * - Display the reason for generating http_responses.
54 * - Log the content length for LOG_LEVEL_CLF correctly
55 * (still incorrect for some fixed responses).
56 * - Reword an incorrect comment about
57 * treat-forbidden-connects-like-blocks violating
59 * - Add some log messages.
61 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
62 * Move mutex initialization into separate
63 * function and exit in case of errors.
65 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
66 * - Start to reduce chat()'s line count and move
67 * parts of it into separate functions.
68 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
70 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
71 * - Add Proxy-Agent header to HTTP snippets that are
72 * supposed to reach HTTP clients only.
73 * - Made a few CONNECT log messages more descriptive.
74 * - Catch completely empty server responses (as seen
75 * with Tor's fake ".noconnect" top level domain).
76 * - Use shiny new "forwarding-failed" template for socks errors.
78 * Revision 1.118 2007/01/07 07:43:43 joergs
79 * AmigaOS4 support added.
81 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
82 * Added config option accept-intercepted-requests
83 * and disabled it by default.
85 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
86 * Reverted parts of my last commit
87 * to keep error handling working.
89 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
90 * Fixed gcc43 conversion warnings.
92 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
93 * Fix -pedantic ISO C warning about converting
94 * from function pointer to object pointer.
96 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
97 * Silence compiler warning I introduced with my last commit.
99 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
100 * Mutex protect rand() if POSIX threading
101 * is used, warn the user if that's not possible
102 * and stop using it on _WIN32 where it could
105 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
106 * Don't prevent core dumps by catching SIGABRT.
107 * It's rude and makes debugging unreasonable painful.
109 * Revision 1.110 2006/12/13 14:52:53 etresoft
110 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
112 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
113 * Privoxy is now able to run as intercepting
114 * proxy in combination with any packet filter
115 * that does the port redirection. The destination
116 * is extracted from the "Host:" header which
117 * should be available for nearly all requests.
119 * Moved HTTP snipplets into jcc.c.
120 * Added error message for gopher proxy requests.
122 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
123 * Only unlink the pidfile if it's actually used.
125 * Change order of interception checks to make
126 * it possible to block or redirect requests for
129 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
130 * Make pthread mutex locking more generic. Instead of
131 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
132 * and use mutex locking unless there is an _r function
133 * available. Better safe than sorry.
135 * Fixes "./configure --disable-pthread" and should result
136 * in less threading-related problems on pthread-using platforms,
137 * but it still doesn't fix BR#1122404.
139 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
140 * Move pthread.h inclusion from jcc.c to jcc.h.
141 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
143 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
144 * Don't exit after receiving the second SIGHUP on Solaris.
146 * Fixes BR 1052235, but the same problem may exist on other
147 * systems. Once 3.0.6 is out we should use sigset()
148 * where available and see if it breaks anything.
150 * Revision 1.104 2006/09/23 13:26:38 roro
151 * Replace TABs by spaces in source code.
153 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
154 * Fix +redirect{}. Didn't work with -fast-redirects.
156 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
157 * Respond with 400 and a short text message
158 * if the client tries to use Privoxy as FTP proxy.
160 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
161 * Make number of retries in case of forwarded-connect problems
162 * a config file option (forwarded-connect-retries) and use 0 as
165 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
166 * Set random(3) seed.
168 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
169 * Follow the OpenBSD port's lead and protect the resolve
170 * functions on OpenBSD as well.
172 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
173 * --user fix. Only use the user as group if no group is specified.
174 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
176 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
177 * Windows service (re-)integration
181 * --install[:service_name]
182 * --uninstall[:service_name]
185 * They work as follows:
186 * --install will create a service for you and then terminate.
187 * By default the service name will be "privoxy" (without the quotes).
188 * However you can run multiple services if you wish, just by adding
189 * a colon and then a name (no spaces).
191 * --uninstall follows the exact same rules a --install.
193 * --service is used when the program is executed by the service
194 * control manager, and in normal circumstances would never be
195 * used as a command line argument.
197 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
198 * Windows service integration
200 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
201 * Incorporate Fabian Keil's patch work:
202 http://www.fabiankeil.de/sourcecode/privoxy/
204 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
205 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
206 * with what was really the latest development (the v_3_0_branch branch)
208 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
209 * Thanks to Jindrich Makovicka for a race condition fix for the log
210 * file. The race condition remains for non-pthread implementations.
211 * Reference patch #1175720.
213 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
214 * Add the ability to check jpeg images for invalid
215 * lengths of comment blocks. Defensive strategy
216 * against the exploit:
217 * Microsoft Security Bulletin MS04-028
218 * Buffer Overrun in JPEG Processing (GDI+) Could
219 * Allow Code Execution (833987)
220 * Enabled with +inspect-jpegs in actions files.
222 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
223 * - Fixed usage info for non-unix platforms
224 * - Fixed small cmdline parsing bug
226 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
227 * Diagnostics: Now preserve the returncode of pthread_create
228 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
230 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
231 * No longer ignore SIGCHLD. Fixes bug #769381
233 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
234 * Close jarfile on graceful exit, remove stray line
236 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
237 * Cosmetics: Killed a warning, a typo and an allocation left at exit
239 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
240 * No longer rely on non-POSIX.1 extensions of getcwd().
243 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
244 * Replaced setenv() by posix-compliant putenv()
245 * Thanks to Neil McCalden (nmcc AT users.sf.net).
247 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
248 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
250 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
251 * Clean-up and extension of improvements for forked mode:
252 * - Child's return code now consists of flags RC_FLAG_*
253 * - Reporting toggle to parent now properly #ifdef'ed
254 * - Children now report blocking to parent. This enables
255 * statistics in forked mode
257 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
258 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
259 * Children now report having been toggled through _exit(17), parents
260 * watch for that code and toggle themselves if found.
262 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
263 * Wrapping all *_r functions (the non-_r versions of them) with
264 * mutex semaphores for OSX. Hopefully this will take care of all
265 * of those pesky crash reports.
267 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
268 * Fixed two mostly harmless mem leaks
270 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
271 * Fix: Head of global clients list now initialized to NULL
273 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
274 * Added basic support for OPTIONS and TRACE HTTP methods:
275 * - New interceptor direct_response() added in chat().
276 * - sed() moved to earlier in the process, so that the
277 * Host: header is evaluated before actions and forwarding
280 * Revision 1.92 2002/05/08 16:00:46 oes
281 * Chat's buffer handling:
282 * - Fixed bug with unchecked out-of-mem conditions
283 * while reading client request & server headers
284 * - No longer predict if the buffer limit will be exceeded
285 * in the next read -- check add_to_iob's new
286 * return code. If buffer couldn't be extended
287 * (policy or out-of-mem) while
288 * - reading from client: abort
289 * - reading server headers: send error page
290 * - buffering server body for filter: flush,
291 * and if that fails: send error page
293 * Revision 1.91 2002/04/08 20:35:58 swa
296 * Revision 1.90 2002/04/02 14:57:28 oes
297 * Made sending wafers independent of FEATURE_COOKIE_JAR
299 * Revision 1.89 2002/03/31 17:18:59 jongfoster
300 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
302 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
303 * More compiler warning message maintenance
305 * Revision 1.87 2002/03/26 22:29:54 swa
306 * we have a new homepage!
308 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
309 * Workaround for closing the jarfile before load_config() comes around again
311 * Revision 1.85 2002/03/24 15:23:33 jongfoster
314 * Revision 1.84 2002/03/24 13:25:43 swa
315 * name change related issues
317 * Revision 1.83 2002/03/16 23:54:06 jongfoster
318 * Adding graceful termination feature, to help look for memory leaks.
319 * If you enable this (which, by design, has to be done by hand
320 * editing config.h) and then go to http://i.j.b/die, then the program
321 * will exit cleanly after the *next* request. It should free all the
322 * memory that was used.
324 * Revision 1.82 2002/03/13 00:27:05 jongfoster
327 * Revision 1.81 2002/03/12 01:42:50 oes
328 * Introduced modular filters
330 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
331 * OS/2 port maintenance:
332 * - Fixed EMX build - it had decayed a little
333 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
334 * substituted a memset for now.
336 * Revision 1.79 2002/03/09 20:03:52 jongfoster
337 * - Making various functions return int rather than size_t.
338 * (Undoing a recent change). Since size_t is unsigned on
339 * Windows, functions like read_socket that return -1 on
340 * error cannot return a size_t.
342 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
343 * crashes, and also frequently caused JB to jump to 100%
344 * CPU and stay there. (Because it thought it had just
345 * read ((unsigned)-1) == 4Gb of data...)
347 * - The signature of write_socket has changed, it now simply
348 * returns success=0/failure=nonzero.
350 * - Trying to get rid of a few warnings --with-debug on
351 * Windows, I've introduced a new type "jb_socket". This is
352 * used for the socket file descriptors. On Windows, this
353 * is SOCKET (a typedef for unsigned). Everywhere else, it's
354 * an int. The error value can't be -1 any more, so it's
355 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
356 * Windows it maps to the #define INVALID_SOCKET.)
358 * - The signature of bind_port has changed.
360 * Revision 1.78 2002/03/08 21:35:04 oes
361 * Added optional group supplement to --user option. Will now use default group of user if no group given
363 * Revision 1.77 2002/03/07 03:52:06 oes
364 * - Fixed compiler warnings etc
365 * - Improved handling of failed DNS lookups
367 * Revision 1.76 2002/03/06 22:54:35 jongfoster
368 * Automated function-comment nitpicking.
370 * Revision 1.75 2002/03/06 10:02:19 oes
371 * Fixed stupid bug when --user was not given
373 * Revision 1.74 2002/03/06 00:49:31 jongfoster
374 * Fixing warning on Windows
375 * Making #ifdefs that refer to the same variable consistently
376 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
378 * Revision 1.73 2002/03/05 23:57:30 hal9
379 * Stray character 's' on line 1618 was breaking build.
381 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
382 * - Re-enable OS/2 building after new parms were added
383 * - Fix false out of memory report when resolving CGI templates when no IP
384 * address is available of failed attempt (a la no such domain)
386 * Revision 1.71 2002/03/05 18:13:56 oes
387 * Added --user option
389 * Revision 1.70 2002/03/05 04:52:42 oes
390 * Deleted non-errlog debugging code
392 * Revision 1.69 2002/03/04 23:50:00 jongfoster
393 * Splitting off bind_port() call into bind_port_helper(), with
396 * Revision 1.68 2002/03/04 20:17:32 oes
399 * Revision 1.67 2002/03/04 18:18:57 oes
400 * - Removed _DEBUG mode
401 * - Cleand up cmdline parsing
402 * - Introduced --no-daemon, --pidfile options
403 * - Cleaned up signal handling:
404 * - Terminate cleanly on INT, TERM and ABRT
405 * - Schedule logfile for re-opening on HUP
406 * - Ignore CHLD and PIPE
407 * - Leave the rest with their default handlers
408 * - Uniform handler registration
409 * - Added usage() function
410 * - Played styleguide police
412 * Revision 1.66 2002/03/03 15:06:55 oes
413 * Re-enabled automatic config reloading
415 * Revision 1.65 2002/03/03 14:49:11 oes
416 * Fixed CLF logging: Now uses client's original HTTP request
418 * Revision 1.64 2002/03/03 09:18:03 joergs
419 * Made jumbjuster work on AmigaOS again.
421 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
422 * Clean up a little CRLF unpleasantness that suddenly appeared
424 * Revision 1.62 2002/02/20 23:17:23 jongfoster
425 * Detecting some out-of memory conditions and exiting with a log message.
427 * Revision 1.61 2002/01/17 21:01:52 jongfoster
428 * Moving all our URL and URL pattern parsing code to urlmatch.c.
430 * Revision 1.60 2001/12/30 14:07:32 steudten
431 * - Add signal handling (unix)
432 * - Add SIGHUP handler (unix)
433 * - Add creation of pidfile (unix)
434 * - Add action 'top' in rc file (RH)
435 * - Add entry 'SIGNALS' to manpage
436 * - Add exit message to logfile (unix)
438 * Revision 1.59 2001/12/13 14:07:18 oes
439 * Fixed Bug: 503 error page now sent OK
441 * Revision 1.58 2001/11/30 23:37:24 jongfoster
442 * Renaming the Win32 config file to config.txt - this is almost the
443 * same as the corresponding UNIX name "config"
445 * Revision 1.57 2001/11/16 00:47:43 jongfoster
446 * Changing the tty-disconnection code to use setsid().
448 * Revision 1.56 2001/11/13 20:20:54 jongfoster
449 * Tabs->spaces, fixing a bug with missing {} around an if()
451 * Revision 1.55 2001/11/13 20:14:53 jongfoster
452 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
454 * Revision 1.54 2001/11/07 00:03:14 steudten
455 * Give reliable return value if an error
456 * occurs not just 0 with new daemon mode.
458 * Revision 1.53 2001/11/05 21:41:43 steudten
459 * Add changes to be a real daemon just for unix os.
460 * (change cwd to /, detach from controlling tty, set
461 * process group and session leader to the own process.
463 * Add some fatal-error log message for failed malloc().
464 * Add '-d' if compiled with 'configure --with-debug' to
465 * enable debug output.
467 * Revision 1.52 2001/10/26 20:11:20 jongfoster
468 * Fixing type mismatch
470 * Revision 1.51 2001/10/26 17:38:28 oes
473 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
474 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
475 * threads to call select() simultaneously. So, it's time to do a real, live,
476 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
477 * (native). Both versions will work, but using __OS2__ offers multi-threading.
479 * Revision 1.49 2001/10/23 21:41:35 jongfoster
480 * Added call to initialize the (statically-allocated of course)
481 * "out of memory" CGI response.
483 * Revision 1.48 2001/10/10 19:56:46 jongfoster
484 * Moving some code that wasn't cookie-related out of an #ifdef
487 * Revision 1.47 2001/10/10 16:44:36 oes
488 * Added CONNECT destination port limitation check
490 * Revision 1.46 2001/10/08 15:17:41 oes
491 * Re-enabled SSL forwarding
493 * Revision 1.45 2001/10/07 15:42:11 oes
494 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
496 * Moved downgrading of the HTTP version from parse_http_request to
497 * chat(), since we can't decide if it is necessary before we have
498 * determined the actions for the URL. The HTTP command is now
499 * *always* re-built so the repairs need no longer be special-cased.
501 * filter_popups now gets a csp pointer so it can raise the new
502 * CSP_FLAG_MODIFIED flag.
506 * Added configurable size limit for the IOB. If the IOB grows so
507 * large that the next read would exceed the limit, the header
508 * is generated, and the header & unfiltered buffer are flushed
509 * to the client. Chat then continues in non-buffering,
510 * non-filtering body mode.
512 * Revision 1.44 2001/10/02 18:13:57 oes
515 * Revision 1.43 2001/10/02 15:32:13 oes
516 * Moved generation of hdr
518 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
519 * Cleaning up 2 compiler warnings on OS/2.
521 * Revision 1.41 2001/09/16 17:05:14 jongfoster
522 * Removing unused #include showarg.h
524 * Revision 1.40 2001/09/16 15:41:45 jongfoster
525 * Fixing signed/unsigned comparison warning.
527 * Revision 1.39 2001/09/16 13:21:27 jongfoster
528 * Changes to use new list functions.
530 * Revision 1.38 2001/09/16 13:01:46 jongfoster
531 * Removing redundant function call that zeroed zalloc()'d memory.
533 * Revision 1.37 2001/09/10 11:12:24 oes
534 * Deleted unused variable
536 * Revision 1.36 2001/09/10 10:56:15 oes
537 * Silenced compiler warnings
539 * Revision 1.35 2001/07/31 14:44:22 oes
540 * Deleted unused size parameter from filter_popups()
542 * Revision 1.34 2001/07/30 22:08:36 jongfoster
543 * Tidying up #defines:
544 * - All feature #defines are now of the form FEATURE_xxx
545 * - Permanently turned off WIN_GUI_EDIT
546 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
548 * Revision 1.33 2001/07/29 19:32:00 jongfoster
549 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
551 * Revision 1.32 2001/07/29 18:47:05 jongfoster
552 * Adding missing #include "loadcfg.h"
554 * Revision 1.31 2001/07/29 12:17:48 oes
555 * Applied pthread fix by Paul Lieverse
557 * Revision 1.30 2001/07/25 22:57:13 jongfoster
558 * __BEOS__ no longer overrides FEATURE_PTHREAD.
559 * This is because FEATURE_PTHREAD will soon be widely used, so I
560 * want to keep it simple.
562 * Revision 1.29 2001/07/24 12:47:06 oes
563 * Applied BeOS support update by Eugenia
565 * Revision 1.28 2001/07/23 13:26:12 oes
566 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
568 * Revision 1.27 2001/07/19 19:09:47 haroon
569 * - Added code to take care of the situation where while processing the first
570 * server response (which includes the server header), after finding the end
571 * of the headers we were not looking past the end of the headers for
572 * content modification. I enabled it for filter_popups.
573 * Someone else should look to see if other similar operations should be
574 * done to the discarded portion of the buffer.
576 * Note 2001/07/20: No, the other content modification mechanisms will process
577 * the whole iob later anyway. --oes
579 * Revision 1.26 2001/07/18 12:31:36 oes
582 * Revision 1.25 2001/07/15 19:43:49 jongfoster
583 * Supports POSIX threads.
584 * Also removed some unused #includes.
586 * Revision 1.24 2001/07/13 14:00:40 oes
587 * - Generic content modification scheme:
588 * Each feature has its own applicability flag that is set
589 * from csp->action->flags.
590 * Replaced the "filtering" int flag , by a function pointer
591 * "content_filter" to the function that will do the content
592 * modification. If it is != NULL, the document will be buffered
593 * and processed through *content_filter, which must set
594 * csp->content_length and return a modified copy of the body
595 * or return NULL (on failiure).
596 * - Changed csp->is_text to the more generic bitmap csp->content_type
597 * which can currently take the valued CT_TEXT or CT_GIF
599 * - Removed all #ifdef PCRS
601 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
602 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
603 * the `block_acl' call. This prevents a compilation error when the user
604 * does not wish to use the "ACL" feature.
606 * Revision 1.22 2001/06/29 21:45:41 oes
607 * Indentation, CRLF->LF, Tab-> Space
609 * Revision 1.21 2001/06/29 13:29:36 oes
610 * - Cleaned up, improved comments
611 * - Unified all possible interceptors (CGI,
612 * block, trust, fast_redirect) in one
613 * place, with one (CGI) answer generation
614 * mechansim. Much clearer now.
615 * - Removed the GIF image generation, which
616 * is now done in filters.c:block_url()
617 * - Made error conditions like domain lookup
618 * failiure or (various) problems while talking
619 * to the server use cgi.c:error_response()
620 * instead of generating HTML/HTTP in chat() (yuck!)
621 * - Removed logentry from cancelled commit
623 * Revision 1.20 2001/06/09 10:55:28 jongfoster
624 * Changing BUFSIZ ==> BUFFER_SIZE
626 * Revision 1.19 2001/06/07 23:12:52 jongfoster
627 * Replacing function pointer in struct gateway with a directly
628 * called function forwarded_connect().
629 * Replacing struct gateway with struct forward_spec
631 * Revision 1.18 2001/06/03 19:12:16 oes
632 * introduced new cgi handling
634 * Revision 1.17 2001/06/01 20:07:23 jongfoster
635 * Now uses action +image-blocker{} rather than config->tinygif
637 * Revision 1.16 2001/06/01 18:49:17 jongfoster
638 * Replaced "list_share" with "list" - the tiny memory gain was not
639 * worth the extra complexity.
641 * Revision 1.15 2001/05/31 21:24:47 jongfoster
642 * Changed "permission" to "action" throughout.
643 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
644 * Moved vanilla wafer check into chat(), since we must now
645 * decide whether or not to add it based on the URL.
647 * Revision 1.14 2001/05/29 20:14:01 joergs
648 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
651 * Revision 1.13 2001/05/29 09:50:24 jongfoster
652 * Unified blocklist/imagelist/permissionslist.
653 * File format is still under discussion, but the internal changes
656 * Also modified interceptor behaviour:
657 * - We now intercept all URLs beginning with one of the following
658 * prefixes (and *only* these prefixes):
660 * * http://ijbswa.sf.net/config/
661 * * http://ijbswa.sourceforge.net/config/
662 * - New interceptors "home page" - go to http://i.j.b/ to see it.
663 * - Internal changes so that intercepted and fast redirect pages
664 * are not replaced with an image.
665 * - Interceptors now have the option to send a binary page direct
666 * to the client. (i.e. ijb-send-banner uses this)
667 * - Implemented show-url-info interceptor. (Which is why I needed
668 * the above interceptors changes - a typical URL is
669 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
670 * The previous mechanism would not have intercepted that, and
671 * if it had been intercepted then it then it would have replaced
674 * Revision 1.12 2001/05/27 22:17:04 oes
676 * - re_process_buffer no longer writes the modified buffer
677 * to the client, which was very ugly. It now returns the
678 * buffer, which it is then written by chat.
680 * - content_length now adjusts the Content-Length: header
681 * for modified documents rather than crunch()ing it.
682 * (Length info in csp->content_length, which is 0 for
683 * unmodified documents)
685 * - For this to work, sed() is called twice when filtering.
687 * Revision 1.11 2001/05/26 17:27:53 jongfoster
688 * Added support for CLF and fixed LOG_LEVEL_LOG.
689 * Also did CRLF->LF fix of my previous patch.
691 * Revision 1.10 2001/05/26 15:26:15 jongfoster
692 * ACL feature now provides more security by immediately dropping
693 * connections from untrusted hosts.
695 * Revision 1.9 2001/05/26 00:28:36 jongfoster
696 * Automatic reloading of config file.
697 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
698 * Most of the global variables have been moved to a new
699 * struct configuration_spec, accessed through csp->config->globalname
700 * Most of the globals remaining are used by the Win32 GUI.
702 * Revision 1.8 2001/05/25 22:43:18 jongfoster
703 * Fixing minor memory leak and buffer overflow.
705 * Revision 1.7 2001/05/25 22:34:30 jongfoster
708 * Revision 1.6 2001/05/23 00:13:58 joergs
709 * AmigaOS support fixed.
711 * Revision 1.5 2001/05/22 18:46:04 oes
713 * - Enabled filtering banners by size rather than URL
714 * by adding patterns that replace all standard banner
715 * sizes with the "Junkbuster" gif to the re_filterfile
717 * - Enabled filtering WebBugs by providing a pattern
718 * which kills all 1x1 images
720 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
721 * which is selected by the (nonstandard and therefore
722 * capital) letter 'U' in the option string.
723 * It causes the quantifiers to be ungreedy by default.
724 * Appending a ? turns back to greedy (!).
726 * - Added a new interceptor ijb-send-banner, which
727 * sends back the "Junkbuster" gif. Without imagelist or
728 * MSIE detection support, or if tinygif = 1, or the
729 * URL isn't recognized as an imageurl, a lame HTML
730 * explanation is sent instead.
732 * - Added new feature, which permits blocking remote
733 * script redirects and firing back a local redirect
735 * The feature is conditionally compiled, i.e. it
736 * can be disabled with --disable-fast-redirects,
737 * plus it must be activated by a "fast-redirects"
738 * line in the config file, has its own log level
739 * and of course wants to be displayed by show-proxy-args
740 * Note: Boy, all the #ifdefs in 1001 locations and
741 * all the fumbling with configure.in and acconfig.h
742 * were *way* more work than the feature itself :-(
744 * - Because a generic redirect template was needed for
745 * this, tinygif = 3 now uses the same.
747 * - Moved GIFs, and other static HTTP response templates
752 * - Removed some >400 CRs again (Jon, you really worked
755 * Revision 1.4 2001/05/21 19:34:01 jongfoster
756 * Made failure to bind() a fatal error.
758 * Revision 1.3 2001/05/20 01:21:20 jongfoster
759 * Version 2.9.4 checkin.
760 * - Merged popupfile and cookiefile, and added control over PCRS
761 * filtering, in new "permissionsfile".
762 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
763 * file error you now get a message box (in the Win32 GUI) rather
764 * than the program exiting with no explanation.
765 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
767 * - Removed tabs from "config"
768 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
769 * - Bumped up version number.
771 * Revision 1.2 2001/05/17 22:34:44 oes
772 * - Added hint on GIF char array generation to jcc.c
773 * - Cleaned CRLF's from the sources and related files
774 * - Repaired logging for REF and FRC
776 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
777 * Initial import of version 2.9.3 source tree
780 *********************************************************************/
786 #include <sys/types.h>
795 # ifndef FEATURE_PTHREAD
799 # include <windows.h>
800 # include <process.h>
801 # endif /* ndef FEATURE_PTHREAD */
804 # ifndef _WIN_CONSOLE
806 # endif /* ndef _WIN_CONSOLE */
807 # include "w32svrapi.h"
809 #else /* ifndef _WIN32 */
811 # if !defined (__OS2__)
813 # include <sys/wait.h>
814 # endif /* ndef __OS2__ */
815 # include <sys/time.h>
816 # include <sys/stat.h>
817 # include <sys/ioctl.h>
820 #include <sys/termios.h>
831 # include <socket.h> /* BeOS has select() for sockets only. */
832 # include <OS.h> /* declarations for threads and stuff. */
835 # if defined(__EMX__) || defined(__OS2__)
836 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
841 #define bzero(B,N) memset(B,0x00,n)
856 #include "killpopup.h"
857 #include "miscutil.h"
859 #include "jbsockets.h"
864 #include "urlmatch.h"
866 const char jcc_h_rcs[] = JCC_H_VERSION;
867 const char project_h_rcs[] = PROJECT_H_VERSION;
870 struct client_state clients[1];
871 struct file_list files[1];
873 #ifdef FEATURE_STATISTICS
874 int urls_read = 0; /* total nr of urls read inc rejected */
875 int urls_rejected = 0; /* total nr of urls rejected */
876 #endif /* def FEATURE_STATISTICS */
878 #ifdef FEATURE_GRACEFUL_TERMINATION
882 static void listen_loop(void);
883 static void chat(struct client_state *csp);
885 void serve(struct client_state *csp);
886 #else /* ifndef AMIGA */
887 static void serve(struct client_state *csp);
888 #endif /* def AMIGA */
891 static int32 server_thread(void *data);
892 #endif /* def __BEOS__ */
895 #define sleep(N) Sleep(((N) * 1000))
899 #define sleep(N) DosSleep(((N) * 100))
902 #ifdef FEATURE_PTHREAD
903 pthread_mutex_t log_mutex;
904 pthread_mutex_t log_init_mutex;
906 #ifndef HAVE_GMTIME_R
907 pthread_mutex_t gmtime_mutex;
908 #endif /* ndef HAVE_GMTIME_R */
910 #ifndef HAVE_LOCALTIME_R
911 pthread_mutex_t localtime_mutex;
912 #endif /* ndef HAVE_GMTIME_R */
914 #ifndef HAVE_GETHOSTBYADDR_R
915 pthread_mutex_t gethostbyaddr_mutex;
916 #endif /* ndef HAVE_GETHOSTBYADDR_R */
918 #ifndef HAVE_GETHOSTBYNAME_R
919 pthread_mutex_t gethostbyname_mutex;
920 #endif /* ndef HAVE_GETHOSTBYNAME_R */
923 pthread_mutex_t rand_mutex;
924 #endif /* ndef HAVE_RANDOM */
926 #endif /* FEATURE_PTHREAD */
928 #if defined(unix) || defined(__EMX__)
929 const char *basedir = NULL;
930 const char *pidfile = NULL;
931 int received_hup_signal = 0;
932 #endif /* defined unix */
934 /* The vanilla wafer. */
935 static const char VANILLA_WAFER[] =
936 "NOTICE=TO_WHOM_IT_MAY_CONCERN_"
937 "Do_not_send_me_any_copyrighted_information_other_than_the_"
938 "document_that_I_am_requesting_or_any_of_its_necessary_components._"
939 "In_particular_do_not_send_me_any_cookies_that_"
940 "are_subject_to_a_claim_of_copyright_by_anybody._"
941 "Take_notice_that_I_refuse_to_be_bound_by_any_license_condition_"
942 "(copyright_or_otherwise)_applying_to_any_cookie._";
944 /* HTTP snipplets. */
945 const char CSUCCEED[] =
946 "HTTP/1.0 200 Connection established\n"
947 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
949 const char CHEADER[] =
950 "HTTP/1.0 400 Invalid header received from browser\r\n"
951 "Connection: close\r\n\r\n"
952 "Invalid header received from browser.\r\n";
954 const char CFORBIDDEN[] =
955 "HTTP/1.0 403 Connection not allowable\r\n"
956 "Proxy-Agent: Privoxy " VERSION "\r\n"
957 "X-Hint: If you read this message interactively, then you know why this happens ,-)\r\n"
958 "Connection: close\r\n\r\n";
960 const char FTP_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 FTP.\r\n";
965 const char GOPHER_RESPONSE[] =
966 "HTTP/1.0 400 Invalid request received from browser\r\n"
967 "Connection: close\r\n\r\n"
968 "Invalid request. Privoxy doesn't support gopher.\r\n";
970 /* XXX: should be a template */
971 const char MISSING_DESTINATION_RESPONSE[] =
972 "HTTP/1.0 400 Bad request received from browser\r\n"
973 "Proxy-Agent: Privoxy " VERSION "\r\n"
974 "Connection: close\r\n\r\n"
975 "Bad request. Privoxy was unable to extract the destination.\r\n";
977 /* XXX: should be a template */
978 const char NO_SERVER_DATA_RESPONSE[] =
979 "HTTP/1.0 502 Server or forwarder response empty\r\n"
980 "Proxy-Agent: Privoxy " VERSION "\r\n"
981 "Connection: close\r\n\r\n"
982 "Empty server or forwarder response.\r\n"
983 "The connection was closed without sending any data.\r\n";
985 /* XXX: should be a template */
986 const char NULL_BYTE_RESPONSE[] =
987 "HTTP/1.0 400 Bad request received from browser\r\n"
988 "Proxy-Agent: Privoxy " VERSION "\r\n"
989 "Connection: close\r\n\r\n"
990 "Bad request. Null byte(s) before end of request.\r\n";
992 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
993 /*********************************************************************
995 * Function : sig_handler
997 * Description : Signal handler for different signals.
998 * Exit gracefully on TERM and INT
999 * or set a flag that will cause the errlog
1000 * to be reopened by the main thread on HUP.
1003 * 1 : the_signal = the signal cause this function to call
1007 *********************************************************************/
1008 static void sig_handler(int the_signal)
1014 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1025 received_hup_signal = 1;
1030 * We shouldn't be here, unless we catch signals
1031 * in main() that we can't handle here!
1033 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1041 /*********************************************************************
1043 * Function : client_protocol_is_unsupported
1045 * Description : Checks if the client used a known unsupported
1046 * protocol and deals with it by sending an error
1050 * 1 : csp = Current client state (buffers, headers, etc...)
1051 * 2 : req = the first request line send by the client
1053 * Returns : TRUE if an error response has been generated, or
1054 * FALSE if the request doesn't look invalid.
1056 *********************************************************************/
1057 int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1059 char buf[BUFFER_SIZE];
1062 * If it's a FTP or gopher request, we don't support it.
1064 * These checks are better than nothing, but they might
1065 * not work in all configurations and some clients might
1066 * have problems digesting the answer.
1068 * They should, however, never cause more problems than
1069 * Privoxy's old behaviour (returning the misleading HTML
1072 * "Could not resolve http://(ftp|gopher)://example.org").
1074 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1076 if (!strncmpic(req, "GET ftp://", 10))
1078 strcpy(buf, FTP_RESPONSE);
1079 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1080 csp->ip_addr_str, req);
1084 strcpy(buf, GOPHER_RESPONSE);
1085 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1086 csp->ip_addr_str, req);
1088 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1090 write_socket(csp->cfd, buf, strlen(buf));
1099 /*********************************************************************
1101 * Function : get_request_destination_elsewhere
1103 * Description : If the client's request was redirected into
1104 * Privoxy without the client's knowledge,
1105 * the request line lacks the destination host.
1107 * This function tries to get it elsewhere,
1108 * provided accept-intercepted-requests is enabled.
1110 * "Elsewhere" currently only means "Host: header",
1111 * but in the future we may ask the redirecting
1112 * packet filter to look the destination up.
1114 * If the destination stays unknown, an error
1115 * response is send to the client and headers
1116 * are freed so that chat() can return directly.
1119 * 1 : csp = Current client state (buffers, headers, etc...)
1120 * 2 : headers = a header list
1122 * Returns : JB_ERR_OK if the destination is now known, or
1123 * JB_ERR_PARSE if it isn't.
1125 *********************************************************************/
1126 jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1128 char buf[BUFFER_SIZE];
1131 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1133 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1134 " Privoxy isn't configured to accept intercepted requests.",
1135 csp->ip_addr_str, csp->http->cmd);
1136 /* XXX: Use correct size */
1137 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1138 csp->ip_addr_str, csp->http->cmd);
1140 strcpy(buf, CHEADER);
1141 write_socket(csp->cfd, buf, strlen(buf));
1142 destroy_list(headers);
1144 return JB_ERR_PARSE;
1146 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1148 /* Split the domain we just got for pattern matching */
1149 init_domain_components(csp->http);
1155 /* We can't work without destination. Go spread the news.*/
1157 req = list_to_text(headers);
1159 /* XXX: Use correct size */
1160 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1161 csp->ip_addr_str, csp->http->cmd);
1162 log_error(LOG_LEVEL_ERROR,
1163 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1164 csp->ip_addr_str, csp->http->cmd, req);
1167 strcpy(buf, MISSING_DESTINATION_RESPONSE);
1168 write_socket(csp->cfd, buf, strlen(buf));
1169 destroy_list(headers);
1171 return JB_ERR_PARSE;
1174 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1175 * to get the destination IP address, use it as host directly
1176 * or do a reverse DNS lookup first.
1181 /*********************************************************************
1183 * Function : get_server_headers
1185 * Description : Parses server headers in iob and fills them
1186 * into csp->headers so that they can later be
1190 * 1 : csp = Current client state (buffers, headers, etc...)
1192 * Returns : JB_ERR_OK if everything went fine, or
1193 * JB_ERR_PARSE if the headers were incomplete.
1195 *********************************************************************/
1196 jb_err get_server_headers(struct client_state *csp)
1198 int continue_hack_in_da_house = 0;
1201 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1206 * continue hack in da house. Ignore the ending of
1207 * this head and continue enlisting header lines.
1208 * The reason is described below.
1210 enlist(csp->headers, "");
1211 continue_hack_in_da_house = 0;
1214 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1217 * It's a bodyless continue response, don't
1218 * stop header parsing after reaching it's end.
1220 * As a result Privoxy will concatenate the
1221 * next response's head and parse and deliver
1222 * the headers as if they belonged to one request.
1224 * The client will separate them because of the
1225 * empty line between them.
1227 * XXX: What we're doing here is clearly against
1228 * the intended purpose of the continue header,
1229 * and under some conditions (HTTP/1.0 client request)
1230 * it's a standard violation.
1232 * Anyway, "sort of against the spec" is preferable
1233 * to "always getting confused by Continue responses"
1234 * (Privoxy's behaviour before this hack was added)
1236 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1237 continue_hack_in_da_house = 1;
1239 else if (*header == '\0')
1242 * If the header is empty, but the Continue hack
1243 * isn't active, we can assume that we reached the
1244 * end of the buffer before we hit the end of the
1247 * Inform the caller an let it decide how to handle it.
1249 return JB_ERR_PARSE;
1253 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1256 * XXX: Should we quit the request and return a
1257 * out of memory error page instead?
1259 log_error(LOG_LEVEL_ERROR,
1260 "Out of memory while enlisting server headers. %s lost.",
1270 /*********************************************************************
1272 * Function : crunch_reason
1274 * Description : Translates the crunch reason code into a string.
1277 * 1 : rsp = a http_response
1279 * Returns : A string with the crunch reason or an error description.
1281 *********************************************************************/
1282 const char *crunch_reason(const struct http_response *rsp)
1284 char * reason = NULL;
1286 assert(rsp != NULL);
1289 return "Internal error while searching for crunch reason";
1292 switch (rsp->reason)
1294 case RSP_REASON_UNSUPPORTED:
1295 reason = "Unsupported HTTP feature";
1297 case RSP_REASON_BLOCKED:
1300 case RSP_REASON_UNTRUSTED:
1301 reason = "Untrusted";
1303 case RSP_REASON_REDIRECTED:
1304 reason = "Redirected";
1306 case RSP_REASON_CGI_CALL:
1307 reason = "CGI Call";
1309 case RSP_REASON_NO_SUCH_DOMAIN:
1310 reason = "DNS failure";
1312 case RSP_REASON_FORWARDING_FAILED:
1313 reason = "Forwarding failed";
1315 case RSP_REASON_CONNECT_FAILED:
1316 reason = "Connection failure";
1318 case RSP_REASON_OUT_OF_MEMORY:
1319 reason = "Out of memory (may mask other reasons)";
1322 reason = "No reason recorded";
1330 /*********************************************************************
1332 * Function : send_crunch_response
1334 * Description : Delivers already prepared response for
1335 * intercepted requests, logs the interception
1336 * and frees the response.
1339 * 1 : csp = Current client state (buffers, headers, etc...)
1340 * 1 : rsp = Fully prepared response. Will be freed on exit.
1342 * Returns : Nothing.
1344 *********************************************************************/
1345 void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1347 const struct http_request *http = csp->http;
1348 char status_code[4];
1350 assert(rsp != NULL);
1351 assert(rsp->head != NULL);
1356 * Not supposed to happen. If it does
1357 * anyway, treat it as an unknown error.
1359 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1360 /* return code doesn't matter */
1365 /* If rsp is still NULL, we have serious internal problems. */
1366 log_error(LOG_LEVEL_FATAL,
1367 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1371 * Extract the status code from the actual head
1372 * that was send to the client. It is the only
1373 * way to get it right for all requests, including
1374 * the fixed ones for out-of-memory problems.
1376 * A head starts like this: 'HTTP/1.1 200...'
1380 status_code[0] = rsp->head[9];
1381 status_code[1] = rsp->head[10];
1382 status_code[2] = rsp->head[11];
1383 status_code[3] = '\0';
1385 /* Write the answer to the client */
1386 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1387 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1389 /* There is nothing we can do about it. */
1390 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1393 /* Log that the request was crunched and why. */
1394 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1395 http->hostport, http->path, crunch_reason(rsp));
1396 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1397 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1399 /* Clean up and return */
1400 if (cgi_error_memory() != rsp)
1402 free_http_response(rsp);
1408 /*********************************************************************
1410 * Function : request_contains_null_bytes
1412 * Description : Checks for NULL bytes in the request and sends
1413 * an error message to the client if any were found.
1416 * 1 : csp = Current client state (buffers, headers, etc...)
1417 * 2 : buf = Data from the client's request to check.
1418 * 3 : len = The data length.
1420 * Returns : TRUE if the request contained one or more NULL bytes, or
1423 *********************************************************************/
1424 int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1426 size_t c_len; /* Request lenght when treated as C string */
1428 c_len = strlen(buf);
1433 * Null byte(s) found. Log the request,
1434 * return an error response and hang up.
1436 size_t tmp_len = c_len;
1441 * Replace NULL byte(s) with '°' characters
1442 * so the request can be logged as string.
1443 * XXX: Is there a better replacement character?
1446 tmp_len += strlen(buf+tmp_len);
1447 } while (tmp_len < len);
1449 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1450 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1451 log_error(LOG_LEVEL_HEADER,
1452 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1454 strcpy(buf, NULL_BYTE_RESPONSE);
1455 write_socket(csp->cfd, buf, strlen(buf));
1457 /* XXX: Log correct size */
1458 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1467 /*********************************************************************
1471 * Description : Once a connection to the client has been accepted,
1472 * this function is called (via serve()) to handle the
1473 * main business of the communication. When this
1474 * function returns, the caller must close the client
1477 * FIXME: chat is nearly thousand lines long.
1481 * 1 : csp = Current client state (buffers, headers, etc...)
1483 * Returns : On success, the number of bytes written are returned (zero
1484 * indicates nothing was written). On error, -1 is returned,
1485 * and errno is set appropriately. If count is zero and the
1486 * file descriptor refers to a regular file, 0 will be
1487 * returned without causing any other effect. For a special
1488 * file, the results are not portable.
1490 *********************************************************************/
1491 static void chat(struct client_state *csp)
1494 * This next lines are a little ugly, but they simplifies the if statements
1495 * below. Basically if TOGGLE, then we want the if to test if the
1496 * CSP_FLAG_TOGGLED_ON flag ist set, else we don't. And if FEATURE_FORCE_LOAD,
1497 * then we want the if to test for CSP_FLAG_FORCED , else we don't
1499 #ifdef FEATURE_TOGGLE
1500 # define IS_TOGGLED_ON_AND (csp->flags & CSP_FLAG_TOGGLED_ON) &&
1501 #else /* ifndef FEATURE_TOGGLE */
1502 # define IS_TOGGLED_ON_AND
1503 #endif /* ndef FEATURE_TOGGLE */
1504 #ifdef FEATURE_FORCE_LOAD
1505 # define IS_NOT_FORCED_AND !(csp->flags & CSP_FLAG_FORCED) &&
1506 #else /* ifndef FEATURE_FORCE_LOAD */
1507 # define IS_NOT_FORCED_AND
1508 #endif /* def FEATURE_FORCE_LOAD */
1510 #define IS_ENABLED_AND IS_TOGGLED_ON_AND IS_NOT_FORCED_AND
1512 char buf[BUFFER_SIZE];
1520 int ms_iis5_hack = 0;
1521 size_t byte_count = 0;
1522 int forwarded_connect_retries = 0;
1523 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1524 const struct forward_spec * fwd;
1525 struct http_request *http;
1526 int len; /* for buffer sizes (and negative error codes) */
1527 #ifdef FEATURE_KILL_POPUPS
1528 int block_popups; /* bool, 1==will block popups */
1529 int block_popups_now = 0; /* bool, 1==currently blocking popups */
1530 #endif /* def FEATURE_KILL_POPUPS */
1532 int pcrs_filter; /* bool, 1==will filter through pcrs */
1533 int gif_deanimate; /* bool, 1==will deanimate gifs */
1534 int jpeg_inspect; /* bool, 1==will inspect jpegs */
1536 /* Function that does the content filtering for the current request */
1537 char *(*content_filter)() = NULL;
1539 /* Skeleton for HTTP response, if we should intercept the request */
1540 struct http_response *rsp;
1542 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1543 struct list header_list;
1544 struct list *headers = &header_list;
1548 memset(buf, 0, sizeof(buf));
1551 * Read the client's request. Note that since we're not using select() we
1552 * could get blocked here if a client connected, then didn't say anything!
1557 len = read_socket(csp->cfd, buf, sizeof(buf)-1);
1559 if (len <= 0) break; /* error! */
1561 if (request_contains_null_bytes(csp, buf, len))
1563 /* NULL bytes found and dealt with, just hang up. */
1568 * If there is no memory left for buffering the
1569 * request, there is nothing we can do but hang up
1571 if (add_to_iob(csp, buf, len))
1576 req = get_header(csp);
1580 break; /* no HTTP request! */
1585 continue; /* more to come! */
1588 /* Does the request line look invalid? */
1589 if (client_protocol_is_unsupported(csp, req))
1592 * Yes. The request has already been
1593 * answered with a error response, the buffers
1594 * were freed and we're done with chatting.
1599 #ifdef FEATURE_FORCE_LOAD
1601 * If this request contains the FORCE_PREFIX and blocks
1602 * aren't enforced, get rid of it and set the force flag.
1604 if (strstr(req, FORCE_PREFIX))
1606 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1608 log_error(LOG_LEVEL_FORCE,
1609 "Ignored force prefix in request: \"%s\".", req);
1613 strclean(req, FORCE_PREFIX);
1614 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
1615 csp->flags |= CSP_FLAG_FORCED;
1619 #endif /* def FEATURE_FORCE_LOAD */
1621 switch( parse_http_request(req, http, csp) )
1624 log_error(LOG_LEVEL_ERROR, "Out of memory while parsing request.");
1627 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", req);
1635 if (http->cmd == NULL)
1637 strcpy(buf, CHEADER);
1638 write_socket(csp->cfd, buf, strlen(buf));
1639 /* XXX: Use correct size */
1640 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1641 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
1643 free_http_request(http);
1647 /* grab the rest of the client's headers */
1651 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
1653 len = read_socket(csp->cfd, buf, sizeof(buf));
1656 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
1661 * If there is no memory left for buffering the
1662 * request, there is nothing we can do but hang up
1664 if (add_to_iob(csp, buf, len))
1671 if (p == NULL) break;
1678 if (http->host == NULL)
1681 * If we still don't know the request destination,
1682 * the request is invalid or the client uses
1683 * Privoxy without it's knowledge.
1685 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
1688 * Our attempts to get the request destination
1689 * elsewhere failed or Privoxy is configured
1690 * to only accept proxy requests.
1692 * An error response has already been send
1693 * and we're done here.
1699 /* decide how to route the HTTP request */
1701 if ((fwd = forward_url(http, csp)) == NULL)
1703 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
1704 /* Never get here - LOG_LEVEL_FATAL causes program exit */
1707 /* build the http request to send to the server
1708 * we have to do one of the following:
1710 * create = use the original HTTP request to create a new
1711 * HTTP request that has either the path component
1712 * without the http://domainspec (w/path) or the
1713 * full orininal URL (w/url)
1714 * Note that the path and/or the HTTP version may
1715 * have been altered by now.
1717 * connect = Open a socket to the host:port of the server
1718 * and short-circuit server and client socket.
1720 * pass = Pass the request unchanged if forwarding a CONNECT
1721 * request to a parent proxy. Note that we'll be sending
1722 * the CFAIL message ourselves if connecting to the parent
1723 * fails, but we won't send a CSUCCEED message if it works,
1724 * since that would result in a double message (ours and the
1725 * parent's). After sending the request to the parent, we simply
1728 * here's the matrix:
1731 * +--------+--------+
1733 * 0 | create | connect|
1735 * Forwarding +--------+--------+
1737 * 1 | create | pass |
1739 * +--------+--------+
1744 * Determine the actions for this URL
1746 #ifdef FEATURE_TOGGLE
1747 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
1749 /* Most compatible set of actions (i.e. none) */
1750 init_current_action(csp->action);
1753 #endif /* ndef FEATURE_TOGGLE */
1755 url_actions(http, csp);
1760 * Check if a CONNECT request is allowable:
1761 * In the absence of a +limit-connect action, allow only port 443.
1762 * If there is an action, allow whatever matches the specificaton.
1766 if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443)
1767 || (csp->action->flags & ACTION_LIMIT_CONNECT
1768 && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) )
1770 if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS)
1773 * The response may confuse some clients,
1774 * but makes unblocking easier.
1776 log_error(LOG_LEVEL_ERROR, "Marking suspicious CONNECT request from %s for blocking.",
1778 csp->action->flags |= ACTION_BLOCK;
1783 strcpy(buf, CFORBIDDEN);
1784 write_socket(csp->cfd, buf, strlen(buf));
1785 log_error(LOG_LEVEL_CONNECT, "Denying suspicious CONNECT request from %s", csp->ip_addr_str);
1786 log_error(LOG_LEVEL_CLF, "%s - - [%T] \" \" 403 0", csp->ip_addr_str);
1794 * Downgrade http version from 1.1 to 1.0 if +downgrade
1797 if ( (http->ssl == 0)
1798 && (!strcmpic(http->ver, "HTTP/1.1"))
1799 && (csp->action->flags & ACTION_DOWNGRADE))
1802 http->ver = strdup("HTTP/1.0");
1804 if (http->ver == NULL)
1806 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1811 * Save a copy of the original request for logging
1813 http->ocmd = strdup(http->cmd);
1815 if (http->ocmd == NULL)
1817 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
1821 * (Re)build the HTTP request for non-SSL requests.
1822 * If forwarding, use the whole URL, else, use only the path.
1828 http->cmd = strdup(http->gpc);
1829 string_append(&http->cmd, " ");
1831 if (fwd->forward_host)
1833 string_append(&http->cmd, http->url);
1837 string_append(&http->cmd, http->path);
1839 string_append(&http->cmd, " ");
1840 string_append(&http->cmd, http->ver);
1842 if (http->cmd == NULL)
1844 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1846 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", http->cmd);
1848 enlist(csp->headers, http->cmd);
1850 /* Append the previously read headers */
1851 list_append_list_unique(csp->headers, headers);
1852 destroy_list(headers);
1855 * If the user has not supplied any wafers, and the user has not
1856 * told us to suppress the vanilla wafer, then send the vanilla wafer.
1858 if (list_is_empty(csp->action->multi[ACTION_MULTI_WAFER])
1859 && ((csp->action->flags & ACTION_VANILLA_WAFER) != 0))
1861 enlist(csp->action->multi[ACTION_MULTI_WAFER], VANILLA_WAFER);
1865 #ifdef FEATURE_KILL_POPUPS
1866 block_popups = ((csp->action->flags & ACTION_NO_POPUPS) != 0);
1867 #endif /* def FEATURE_KILL_POPUPS */
1869 pcrs_filter = (csp->rlist != NULL) && /* There are expressions to be used */
1870 (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER]));
1872 gif_deanimate = ((csp->action->flags & ACTION_DEANIMATE) != 0);
1874 jpeg_inspect = ((csp->action->flags & ACTION_JPEG_INSPECT) != 0);
1877 * We have a request. Now, check to see if we need to
1878 * intercept it, i.e. If ..
1882 /* We may not forward the request by rfc2616 sect 14.31 */
1883 (NULL != (rsp = direct_response(csp)))
1885 /* or we are enabled and... */
1886 || (IS_ENABLED_AND (
1888 /* ..the request was blocked */
1889 ( NULL != (rsp = block_url(csp)))
1891 /* ..or untrusted */
1892 #ifdef FEATURE_TRUST
1893 || ( NULL != (rsp = trust_url(csp)))
1894 #endif /* def FEATURE_TRUST */
1896 /* ..or a redirect kicked in */
1897 || ( NULL != (rsp = redirect_url(csp)))
1901 * .. or a CGI call was detected and answered.
1903 * This check comes last to give the user the power
1904 * to deny acces to some (or all) of the cgi pages.
1906 || (NULL != (rsp = dispatch_cgi(csp)))
1911 * Deliver, log and free the interception
1912 * response. Afterwards we're done here.
1914 send_crunch_response(csp, rsp);
1915 #ifdef FEATURE_STATISTICS
1916 /* Count as a rejected request */
1917 csp->flags |= CSP_FLAG_REJECTED;
1918 #endif /* def FEATURE_STATISTICS */
1923 hdr = sed(client_patterns, add_client_headers, csp);
1926 /* FIXME Should handle error properly */
1927 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
1930 list_remove_all(csp->headers);
1932 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
1934 if (fwd->forward_host)
1936 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
1937 fwd->forward_host, fwd->forward_port, http->hostport);
1941 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
1944 /* here we connect to the server, gateway, or the forwarder */
1946 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
1947 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
1949 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
1950 forwarded_connect_retries, http->hostport);
1953 if (csp->sfd == JB_INVALID_SOCKET)
1955 if (fwd->type != SOCKS_NONE)
1958 rsp = error_response(csp, "forwarding-failed", errno);
1960 else if (errno == EINVAL)
1962 rsp = error_response(csp, "no-such-domain", errno);
1966 rsp = error_response(csp, "connect-failed", errno);
1967 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
1972 /* Write the answer to the client */
1975 send_crunch_response(csp, rsp);
1982 if (fwd->forward_host || (http->ssl == 0))
1984 /* write the client's (modified) header to the server
1985 * (along with anything else that may be in the buffer)
1988 if (write_socket(csp->sfd, hdr, strlen(hdr))
1989 || (flush_socket(csp->sfd, csp) < 0))
1991 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
1994 rsp = error_response(csp, "connect-failed", errno);
1998 send_crunch_response(csp, rsp);
2008 * We're running an SSL tunnel and we're not forwarding,
2009 * so just send the "connect succeeded" message to the
2010 * client, flush the rest, and get out of the way.
2012 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 0",
2013 csp->ip_addr_str, http->ocmd);
2015 if (write_socket(csp->cfd, CSUCCEED, sizeof(CSUCCEED)-1))
2023 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2025 /* we're finished with the client's header */
2028 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2030 /* pass data between the client and server
2031 * until one or the other shuts down the connection.
2040 * FD_ZERO here seems to point to an errant macro which crashes.
2041 * So do this by hand for now...
2043 memset(&rfds,0x00,sizeof(fd_set));
2047 FD_SET(csp->cfd, &rfds);
2048 FD_SET(csp->sfd, &rfds);
2050 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2054 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2058 /* this is the body of the browser's request
2059 * just read it and write it.
2062 if (FD_ISSET(csp->cfd, &rfds))
2064 len = read_socket(csp->cfd, buf, sizeof(buf));
2068 break; /* "game over, man" */
2071 if (write_socket(csp->sfd, buf, (size_t)len))
2073 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2080 * The server wants to talk. It could be the header or the body.
2081 * If `hdr' is null, then it's the header otherwise it's the body.
2082 * FIXME: Does `hdr' really mean `host'? No.
2086 if (FD_ISSET(csp->sfd, &rfds))
2089 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2093 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2095 if (http->ssl && (fwd->forward_host == NULL))
2098 * Just hang up. We already confirmed the client's CONNECT
2099 * request with status code 200 and unencrypted content is
2100 * no longer welcome.
2102 log_error(LOG_LEVEL_ERROR,
2103 "CONNECT already confirmed. Unable to tell the client about the problem.");
2107 rsp = error_response(csp, "connect-failed", errno);
2111 send_crunch_response(csp, rsp);
2117 /* Add a trailing zero. This lets filter_popups
2118 * use string operations.
2122 #ifdef FEATURE_KILL_POPUPS
2123 /* Filter the popups on this read. */
2124 if (block_popups_now)
2126 filter_popups(buf, csp);
2128 #endif /* def FEATURE_KILL_POPUPS */
2130 /* Normally, this would indicate that we've read
2131 * as much as the server has sent us and we can
2132 * close the client connection. However, Microsoft
2133 * in its wisdom has released IIS/5 with a bug that
2134 * prevents it from sending the trailing \r\n in
2135 * a 302 redirect header (and possibly other headers).
2136 * To work around this if we've haven't parsed
2137 * a full header we'll append a trailing \r\n
2138 * and see if this now generates a valid one.
2140 * This hack shouldn't have any impacts. If we've
2141 * already transmitted the header or if this is a
2142 * SSL connection, then we won't bother with this
2143 * hack. So we only work on partially received
2144 * headers. If we append a \r\n and this still
2145 * doesn't generate a valid header, then we won't
2146 * transmit anything to the client.
2151 if (server_body || http->ssl)
2154 * If we have been buffering up the document,
2155 * now is the time to apply content modification
2156 * and send the result to the client.
2161 * If the content filter fails, use the original
2162 * buffer and length.
2163 * (see p != NULL ? p : csp->iob->cur below)
2165 if (NULL == (p = (*content_filter)(csp)))
2167 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2170 hdr = sed(server_patterns_light, NULL, csp);
2174 /* FIXME Should handle error properly */
2175 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2178 if (write_socket(csp->cfd, hdr, strlen(hdr))
2179 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2181 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2191 break; /* "game over, man" */
2195 * This is NOT the body, so
2196 * Let's pretend the server just sent us a blank line.
2198 snprintf(buf, sizeof(buf), "\r\n");
2199 len = (int)strlen(buf);
2202 * Now, let the normal header parsing algorithm below do its
2203 * job. If it fails, we'll exit instead of continuing.
2210 * If this is an SSL connection or we're in the body
2211 * of the server document, just write it to the client,
2212 * unless we need to buffer the body for later content-filtering
2215 if (server_body || http->ssl)
2220 * If there is no memory left for buffering the content, or the buffer limit
2221 * has been reached, switch to non-filtering mode, i.e. make & write the
2222 * header, flush the iob and buf, and get out of the way.
2224 if (add_to_iob(csp, buf, len))
2229 log_error(LOG_LEVEL_ERROR, "Flushing header and buffers. Stepping back from filtering.");
2231 hdr = sed(server_patterns, add_server_headers, csp);
2235 * Memory is too tight to even generate the header.
2236 * Send our static "Out-of-memory" page.
2238 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2239 rsp = cgi_error_memory();
2240 send_crunch_response(csp, rsp);
2245 hdrlen = strlen(hdr);
2247 if (write_socket(csp->cfd, hdr, hdrlen)
2248 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
2249 || (write_socket(csp->cfd, buf, (size_t)len)))
2251 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2257 byte_count += hdrlen + (size_t)flushed + (size_t)len;
2259 content_filter = NULL;
2266 if (write_socket(csp->cfd, buf, (size_t)len))
2268 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2272 byte_count += (size_t)len;
2277 /* we're still looking for the end of the
2278 * server's header ... (does that make header
2279 * parsing an "out of body experience" ?
2283 * buffer up the data we just read. If that fails,
2284 * there's little we can do but send our static
2285 * out-of-memory page.
2287 if (add_to_iob(csp, buf, len))
2289 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2290 rsp = cgi_error_memory();
2291 send_crunch_response(csp, rsp);
2296 /* Convert iob into something sed() can digest */
2297 if (JB_ERR_PARSE == get_server_headers(csp))
2301 /* Well, we tried our MS IIS/5
2302 * hack and it didn't work.
2303 * The header is incomplete
2304 * and there isn't anything
2305 * we can do about it.
2311 /* Since we have to wait for
2312 * more from the server before
2313 * we can parse the headers
2314 * we just continue here.
2320 /* Did we actually get anything? */
2321 if (NULL == csp->headers->first)
2323 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2324 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2325 strcpy(buf, NO_SERVER_DATA_RESPONSE);
2326 write_socket(csp->cfd, buf, strlen(buf));
2327 free_http_request(http);
2331 /* we have now received the entire header.
2332 * filter it and send the result to the client
2335 hdr = sed(server_patterns, add_server_headers, csp);
2338 /* FIXME Should handle error properly */
2339 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2342 #ifdef FEATURE_KILL_POPUPS
2343 /* Start blocking popups if appropriate. */
2345 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2346 !http->ssl && /* We talk plaintext */
2347 block_popups) /* Policy allows */
2349 block_popups_now = 1;
2351 * Filter the part of the body that came in the same read
2352 * as the last headers:
2354 filter_popups(csp->iob->cur, csp);
2357 #endif /* def FEATURE_KILL_POPUPS */
2359 /* Buffer and pcrs filter this if appropriate. */
2361 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2362 !http->ssl && /* We talk plaintext */
2363 pcrs_filter) /* Policy allows */
2365 content_filter = pcrs_filter_response;
2368 /* Buffer and gif_deanimate this if appropriate. */
2370 if ((csp->content_type & CT_GIF) && /* It's an image/gif MIME-Type */
2371 !http->ssl && /* We talk plaintext */
2372 gif_deanimate) /* Policy allows */
2374 content_filter = gif_deanimate_response;
2377 /* Buffer and jpg_inspect this if appropriate. */
2379 if ((csp->content_type & CT_JPEG) && /* It's an image/jpeg MIME-Type */
2380 !http->ssl && /* We talk plaintext */
2381 jpeg_inspect) /* Policy allows */
2383 content_filter = jpeg_inspect_response;
2387 * Only write if we're not buffering for content modification
2389 if (!content_filter)
2391 /* write the server's (modified) header to
2392 * the client (along with anything else that
2393 * may be in the buffer)
2396 if (write_socket(csp->cfd, hdr, strlen(hdr))
2397 || ((len = flush_socket(csp->cfd, csp)) < 0))
2399 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2401 /* the write failed, so don't bother
2402 * mentioning it to the client...
2403 * it probably can't hear us anyway.
2409 byte_count += (size_t)len;
2412 /* we're finished with the server's header */
2417 /* If this was a MS IIS/5 hack then it means
2418 * the server has already closed the
2419 * connection. Nothing more to read. Time
2430 return; /* huh? we should never get here */
2433 if (csp->content_length == 0)
2436 * If Privoxy didn't recalculate the
2437 * Content-Lenght, byte_count is still
2440 csp->content_length = byte_count;
2443 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2444 csp->ip_addr_str, http->ocmd, csp->content_length);
2448 /*********************************************************************
2452 * Description : This is little more than chat. We only "serve" to
2453 * to close any socket that chat may have opened.
2456 * 1 : csp = Current client state (buffers, headers, etc...)
2460 *********************************************************************/
2462 void serve(struct client_state *csp)
2463 #else /* ifndef AMIGA */
2464 static void serve(struct client_state *csp)
2465 #endif /* def AMIGA */
2468 close_socket(csp->cfd);
2470 if (csp->sfd != JB_INVALID_SOCKET)
2472 close_socket(csp->sfd);
2475 csp->flags &= ~CSP_FLAG_ACTIVE;
2481 /*********************************************************************
2483 * Function : server_thread
2485 * Description : We only exist to call `serve' in a threaded environment.
2488 * 1 : data = Current client state (buffers, headers, etc...)
2490 * Returns : Always 0.
2492 *********************************************************************/
2493 static int32 server_thread(void *data)
2495 serve((struct client_state *) data);
2502 /*********************************************************************
2506 * Description : Print usage info & exit.
2508 * Parameters : Pointer to argv[0] for identifying ourselves
2512 *********************************************************************/
2513 void usage(const char *myname)
2515 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2517 "Usage: %s [--help] [--version] [configfile]\n"
2519 "Usage: %s [--help] [--version] [--no-daemon] [--pidfile pidfile] [--user user[.group]] [configfile]\n"
2521 "Aborting.\n", myname);
2528 /*********************************************************************
2530 * Function : initialize_mutexes
2532 * Description : Prepares mutexes if mutex support is available.
2536 * Returns : Void, exits in case of errors.
2538 *********************************************************************/
2539 void initialize_mutexes()
2543 #ifdef FEATURE_PTHREAD
2545 * Prepare global mutex semaphores
2547 err = pthread_mutex_init(&log_mutex, 0);
2549 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2552 * XXX: The assumptions below are a bit naive
2553 * and can cause locks that aren't necessary.
2555 * For example older FreeBSD versions (< 6.x?)
2556 * have no gethostbyname_r, but gethostbyname is
2559 #ifndef HAVE_GMTIME_R
2560 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2561 #endif /* ndef HAVE_GMTIME_R */
2563 #ifndef HAVE_LOCALTIME_R
2564 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2565 #endif /* ndef HAVE_GMTIME_R */
2567 #ifndef HAVE_GETHOSTBYADDR_R
2568 if (!err) err = pthread_mutex_init(&gethostbyaddr_mutex, 0);
2569 #endif /* ndef HAVE_GETHOSTBYADDR_R */
2571 #ifndef HAVE_GETHOSTBYNAME_R
2572 if (!err) err = pthread_mutex_init(&gethostbyname_mutex, 0);
2573 #endif /* ndef HAVE_GETHOSTBYNAME_R */
2576 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2577 #endif /* ndef HAVE_RANDOM */
2578 #endif /* FEATURE_PTHREAD */
2581 * TODO: mutex support for mingw32 would be swell.
2586 printf("Fatal error. Mutex initialization failed: %s.\n",
2595 /*********************************************************************
2599 * Description : Load the config file and start the listen loop.
2600 * This function is a lot more *sane* with the `load_config'
2601 * and `listen_loop' functions; although it stills does
2602 * a *little* too much for my taste.
2605 * 1 : argc = Number of parameters (including $0).
2606 * 2 : argv = Array of (char *)'s to the parameters.
2608 * Returns : 1 if : can't open config file, unrecognized directive,
2609 * stats requested in multi-thread mode, can't open the
2610 * log file, can't open the jar file, listen port is invalid,
2611 * any load fails, and can't bind port.
2613 * Else main never returns, the process must be signaled
2614 * to terminate execution. Or, on Windows, use the
2615 * "File", "Exit" menu option.
2617 *********************************************************************/
2619 int real_main(int argc, const char *argv[])
2621 int main(int argc, const char *argv[])
2625 unsigned int random_seed;
2627 struct passwd *pw = NULL;
2628 struct group *grp = NULL;
2637 #if !defined(_WIN32)
2645 * Parse the command line arguments
2647 while (++argc_pos < argc)
2650 /* Check to see if the service must be installed or uninstalled */
2651 if (strncmp(argv[argc_pos], "--install", 9) == 0)
2653 const char *pName = argv[argc_pos] + 9;
2656 exit( (install_service(pName)) ? 0 : 1 );
2658 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
2660 const char *pName = argv[argc_pos] + 11;
2663 exit((uninstall_service(pName)) ? 0 : 1);
2665 else if (strcmp(argv[argc_pos], "--service" ) == 0)
2667 bRunAsService = TRUE;
2668 w32_set_service_cwd();
2669 atexit(w32_service_exit_notify);
2672 #endif /* defined(_WIN32) */
2675 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2677 if (strcmp(argv[argc_pos], "--help") == 0)
2682 else if(strcmp(argv[argc_pos], "--version") == 0)
2684 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
2690 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
2695 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
2697 if (++argc_pos == argc) usage(argv[0]);
2698 pidfile = strdup(argv[argc_pos]);
2701 else if (strcmp(argv[argc_pos], "--user" ) == 0)
2703 if (++argc_pos == argc) usage(argv[argc_pos]);
2705 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
2708 if (NULL == (grp = getgrnam(p)))
2710 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
2714 if (NULL == (pw = getpwnam(argv[argc_pos])))
2716 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
2719 if (p != NULL) *--p = '\0';
2722 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
2727 #endif /* defined(unix) */
2728 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
2730 configfile = argv[argc_pos];
2733 } /* -END- while (more arguments) */
2736 if ( *configfile != '/' )
2738 char *abs_file, cwd[1024];
2740 /* make config-filename absolute here */
2741 if ( !(getcwd(cwd, sizeof(cwd))))
2743 perror("get working dir failed");
2747 if (!(basedir = strdup(cwd))
2748 || (!(abs_file = malloc( strlen( basedir ) + strlen( configfile ) + 5 ))))
2750 perror("malloc failed");
2753 strcpy( abs_file, basedir );
2754 strcat( abs_file, "/" );
2755 strcat( abs_file, configfile );
2756 configfile = abs_file;
2758 #endif /* defined unix */
2762 clients->next = NULL;
2766 #elif defined(_WIN32)
2770 /* Prepare mutexes if supported and necessary. */
2771 initialize_mutexes();
2773 random_seed = (unsigned int)time(NULL);
2775 srandom(random_seed);
2778 #endif /* ifdef HAVE_RANDOM */
2781 * Unix signal handling
2783 * Catch the abort, interrupt and terminate signals for a graceful exit
2784 * Catch the hangup signal so the errlog can be reopened.
2785 * Ignore the broken pipe signals (FIXME: Why?)
2787 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
2790 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
2791 const int ignored_signals[] = { SIGPIPE, 0 };
2793 for (idx = 0; catched_signals[idx] != 0; idx++)
2795 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
2796 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
2798 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
2799 #endif /* ifdef sun */
2801 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
2805 for (idx = 0; ignored_signals[idx] != 0; idx++)
2807 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
2809 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
2814 #else /* ifdef _WIN32 */
2815 # ifdef _WIN_CONSOLE
2817 * We *are* in a windows console app.
2818 * Print a verbose messages about FAQ's and such
2820 printf("%s", win32_blurb);
2821 # endif /* def _WIN_CONSOLE */
2822 #endif /* def _WIN32 */
2825 /* Initialize the CGI subsystem */
2826 cgi_init_error_messages();
2829 * If runnig on unix and without the --nodaemon
2830 * option, become a daemon. I.e. fork, detach
2831 * from tty and get process group leadership
2844 if ( pid < 0 ) /* error */
2849 else if ( pid != 0 ) /* parent */
2854 * must check for errors
2855 * child died due to missing files aso
2858 wpid = waitpid( pid, &status, WNOHANG );
2867 /* Should be more portable, but not as well tested */
2872 #else /* ndef __FreeBSD__ */
2874 #endif /* ndef __FreeBSD__ */
2875 fd = open("/dev/tty", O_RDONLY);
2878 /* no error check here */
2879 ioctl( fd, TIOCNOTTY,0 );
2883 /* FIXME: should close stderr (fd 2) here too, but the test
2885 * and load config file is done in listen_loop() and puts
2886 * some messages on stderr there.
2893 } /* -END- if (!no_daemon) */
2896 * As soon as we have written the PID file, we can switch
2897 * to the user and group ID indicated by the --user option
2903 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
2905 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
2911 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
2913 if (chroot(pw->pw_dir) < 0)
2915 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
2919 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
2922 if (setuid(pw->pw_uid))
2924 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
2928 char putenv_dummy[64];
2930 strcpy(putenv_dummy, "HOME=/");
2931 if (putenv(putenv_dummy) != 0)
2933 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
2936 snprintf(putenv_dummy, 64, "USER=%s", pw->pw_name);
2937 if (putenv(putenv_dummy) != 0)
2939 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
2945 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
2948 #endif /* defined unix */
2951 /* This will be FALSE unless the command line specified --service
2955 /* Yup, so now we must attempt to establish a connection
2956 * with the service dispatcher. This will only work if this
2957 * process was launched by the service control manager to
2958 * actually run as a service. If this isn't the case, i've
2959 * known it take around 30 seconds or so for the call to return.
2962 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
2963 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
2965 /* Service has run, and at this point is now being stopped, so just return */
2970 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
2972 /* An error occurred. Usually it's because --service was wrongly specified
2973 * and we were unable to connect to the Service Control Dispatcher because
2974 * it wasn't expecting us and is therefore not listening.
2976 * For now, just continue below to call the listen_loop function.
2979 #endif /* def _WIN32 */
2989 /*********************************************************************
2991 * Function : bind_port_helper
2993 * Description : Bind the listen port. Handles logging, and aborts
2997 * 1 : config = Privoxy configuration. Specifies port
3000 * Returns : Port that was opened.
3002 *********************************************************************/
3003 static jb_socket bind_port_helper(struct configuration_spec * config)
3008 if ( (config->haddr != NULL)
3009 && (config->haddr[0] == '1')
3010 && (config->haddr[1] == '2')
3011 && (config->haddr[2] == '7')
3012 && (config->haddr[3] == '.') )
3014 log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only",
3017 else if (config->haddr == NULL)
3019 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3024 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3025 config->hport, config->haddr);
3028 result = bind_port(config->haddr, config->hport, &bfd);
3035 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3036 "There may be another Privoxy or some other "
3037 "proxy running on port %d",
3038 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3039 config->hport, config->hport);
3042 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3043 "The hostname is not resolvable",
3044 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3047 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3048 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3051 /* shouldn't get here */
3052 return JB_INVALID_SOCKET;
3055 config->need_bind = 0;
3062 /* Without this simple workaround we get this compiler warning from _beginthread
3063 * warning C4028: formal parameter 1 different from declaration
3065 void w32_service_listen_loop(void *p)
3069 #endif /* def _WIN32 */
3072 /*********************************************************************
3074 * Function : listen_loop
3076 * Description : bind the listen port and enter a "FOREVER" listening loop.
3082 *********************************************************************/
3083 static void listen_loop(void)
3085 struct client_state *csp = NULL;
3087 struct configuration_spec * config;
3089 config = load_config();
3091 bfd = bind_port_helper(config);
3093 #ifdef FEATURE_GRACEFUL_TERMINATION
3094 while (!g_terminate)
3099 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3100 while (waitpid(-1, NULL, WNOHANG) > 0)
3102 /* zombie children */
3104 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3107 * Free data that was used by died threads
3113 * Re-open the errlog after HUP signal
3115 if (received_hup_signal)
3117 init_error_log(Argv[0], config->logfile, config->debug);
3118 received_hup_signal = 0;
3123 #ifdef FEATURE_COOKIE_JAR
3125 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3126 * already open. I think unload_configfile() is not being run, which should do
3127 * this work. Until that can get resolved, we'll use this workaround.
3131 if (csp->config->jar)
3133 fclose(csp->config->jar);
3134 csp->config->jar = NULL;
3136 #endif /* FEATURE_COOKIE_JAR */
3137 #endif /* __OS2__ */
3139 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3141 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3145 csp->flags |= CSP_FLAG_ACTIVE;
3146 csp->sfd = JB_INVALID_SOCKET;
3148 csp->config = config = load_config();
3150 if ( config->need_bind )
3153 * Since we were listening to the "old port", we will not see
3154 * a "listen" param change until the next IJB request. So, at
3155 * least 1 more request must be made for us to find the new
3156 * setting. I am simply closing the old socket and binding the
3159 * Which-ever is correct, we will serve 1 more page via the
3160 * old settings. This should probably be a "show-proxy-args"
3161 * request. This should not be a so common of an operation
3162 * that this will hurt people's feelings.
3167 bfd = bind_port_helper(config);
3170 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3172 if (!accept_connection(csp, bfd))
3174 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3187 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3190 #ifdef FEATURE_TOGGLE
3191 if (global_toggle_state)
3193 csp->flags |= CSP_FLAG_TOGGLED_ON;
3195 #endif /* def FEATURE_TOGGLE */
3197 if (run_loader(csp))
3199 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3200 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3204 if (block_acl(NULL,csp))
3206 log_error(LOG_LEVEL_CONNECT, "Connection dropped due to ACL");
3207 close_socket(csp->cfd);
3211 #endif /* def FEATURE_ACL */
3213 /* add it to the list of clients */
3214 csp->next = clients->next;
3215 clients->next = csp;
3217 if (config->multi_threaded)
3221 /* this is a switch () statment in the C preprocessor - ugh */
3222 #undef SELECTED_ONE_OPTION
3224 /* Use Pthreads in preference to native code */
3225 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3226 #define SELECTED_ONE_OPTION
3228 pthread_t the_thread;
3229 pthread_attr_t attrs;
3231 pthread_attr_init(&attrs);
3232 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3233 errno = pthread_create(&the_thread, &attrs,
3234 (void * (*)(void *))serve, csp);
3235 child_id = errno ? -1 : 0;
3236 pthread_attr_destroy(&attrs);
3240 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3241 #define SELECTED_ONE_OPTION
3242 child_id = _beginthread(
3243 (void (*)(void *))serve,
3248 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3249 #define SELECTED_ONE_OPTION
3250 child_id = _beginthread(
3251 (void(* _Optlink)(void*))serve,
3257 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3258 #define SELECTED_ONE_OPTION
3260 thread_id tid = spawn_thread
3261 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3263 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3265 child_id = (int) tid;
3274 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3275 #define SELECTED_ONE_OPTION
3276 csp->cfd = ReleaseSocket(csp->cfd, -1);
3279 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3280 NP_Output, Output(),
3281 NP_CloseOutput, FALSE,
3282 NP_Name, (ULONG)"privoxy child",
3286 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3287 NP_Output, Output(),
3288 NP_CloseOutput, FALSE,
3289 NP_Name, (ULONG)"privoxy child",
3290 NP_StackSize, 200*1024,
3296 ((struct Task *)child_id)->tc_UserData = csp;
3297 Signal((struct Task *)child_id, SIGF_SINGLE);
3302 #if !defined(SELECTED_ONE_OPTION)
3305 /* This block is only needed when using fork().
3306 * When using threads, the server thread was
3307 * created and run by the call to _beginthread().
3309 if (child_id == 0) /* child */
3312 #ifdef FEATURE_TOGGLE
3313 int inherited_toggle_state = global_toggle_state;
3314 #endif /* def FEATURE_TOGGLE */
3319 * If we've been toggled or we've blocked the request, tell Mom
3322 #ifdef FEATURE_TOGGLE
3323 if (inherited_toggle_state != global_toggle_state)
3325 rc |= RC_FLAG_TOGGLED;
3327 #endif /* def FEATURE_TOGGLE */
3329 #ifdef FEATURE_STATISTICS
3330 if (csp->flags & CSP_FLAG_REJECTED)
3332 rc |= RC_FLAG_BLOCKED;
3334 #endif /* ndef FEATURE_STATISTICS */
3338 else if (child_id > 0) /* parent */
3340 /* in a fork()'d environment, the parent's
3341 * copy of the client socket and the CSP
3345 #if !defined(_WIN32) && !defined(__CYGWIN__)
3347 wait( &child_status );
3350 * Evaluate child's return code: If the child has
3351 * - been toggled, toggle ourselves
3352 * - blocked its request, bump up the stats counter
3355 #ifdef FEATURE_TOGGLE
3356 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3358 global_toggle_state = !global_toggle_state;
3360 #endif /* def FEATURE_TOGGLE */
3362 #ifdef FEATURE_STATISTICS
3364 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3368 #endif /* def FEATURE_STATISTICS */
3370 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3371 close_socket(csp->cfd);
3372 csp->flags &= ~CSP_FLAG_ACTIVE;
3376 #undef SELECTED_ONE_OPTION
3377 /* end of cpp switch () */
3379 if (child_id < 0) /* failed */
3381 char buf[BUFFER_SIZE];
3383 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3385 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3387 write_socket(csp->cfd, buf, strlen(buf));
3388 close_socket(csp->cfd);
3389 csp->flags &= ~CSP_FLAG_ACTIVE;
3400 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3402 /* Clean up. Aim: free all memory (no leaks) */
3403 #ifdef FEATURE_GRACEFUL_TERMINATION
3405 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3407 unload_current_config_file();
3408 unload_current_actions_file();
3409 unload_current_re_filterfile();
3410 #ifdef FEATURE_TRUST
3411 unload_current_trust_file();
3414 if (config->multi_threaded)
3421 } while ((clients->next != NULL) && (--i > 0));
3425 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3436 #ifdef FEATURE_COOKIE_JAR
3437 if (NULL != config->jar)
3439 fclose(config->jar);
3443 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3444 /* Cleanup - remove taskbar icon etc. */
3449 #endif /* FEATURE_GRACEFUL_TERMINATION */