1 const char jcc_rcs[] = "$Id: jcc.c,v 1.150 2007/09/28 16:39:29 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 1.150 2007/09/28 16:39:29 fabiankeil
37 * Execute content filters through execute_content_filter().
39 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
40 * Initialize req to NULL to make sure it's defined if the
41 * first read_socket() call fails. Reported by icmp30.
43 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
44 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
45 * extensive comments moved to user manual.
47 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
48 * Don't crash if a broken header filter wiped out the request line.
50 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
51 * Fix byte_count calculation in case of flushes
52 * and don't parse the server headers a second time.
54 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
55 * - If there's a connection problem after we already forwarded
56 * parts of the original content, just hang up. Fixes BR#1776724.
57 * - Fix warnings about unused code on mingw32.
58 * - In case of flushes, calculate the byte count
59 * less incorrectly (I think).
61 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
62 * Add some more prototypes for static functions.
64 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
65 * Comment out request_contains_null_bytes() until it's used again.
67 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
68 * #1763173 from Stefan Huehner: s@const static@static const@
69 * and declare some more functions static.
71 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
72 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
73 * and explain why they were rejected in the first place.
74 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
75 * CONNECT requests. The request line was missing.
76 * - Add two more XXX reminders as we don't have enough already.
78 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
79 * As Hal noticed, checking dispatch_cgi() as the last cruncher
80 * looks like a bug if CGI requests are blocked unintentionally,
81 * so don't do it unless the user enabled the new config option
82 * "allow-cgi-request-crunching".
84 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
85 * - Allow to rewrite the request destination behind the client's back.
86 * - Turn the weird-looking unconditional for loop that
87 * reads the client request into a conditional while loop.
88 * Move the stuff that only runs once out of the loop.
89 * - Move parts of chat(), server_content_type() and the
90 * necessary stuff to fix BR#1750917 into get_filter_function().
92 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
93 * Temporary workaround for BR#1730105.
95 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
96 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
97 * deadlocks and crashes on OpenBSD and possibly other OS with neither
98 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
99 * Thanks to Ralf Horstmann for report and solution.
101 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
102 * Add forward-override{} to change the forwarding settings through
103 * action sections. This is mainly interesting to forward different
104 * clients differently (for example based on User-Agent or request
107 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
108 * - Let usage() mention the --chroot parameter.
109 * - Use read_socket() consistently and always leave
110 * the last buffer byte alone, even in cases where
111 * null termination (currently) doesn't matter.
113 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
114 * - Fix config file loading on Unix if no config file is specified.
115 * Since r1.97 Privoxy would always interpret the last argument as
116 * config file, even if it's a valid command line option.
117 * - Abort in case of unrecognized command line options. Closes #1719696.
118 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
119 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
121 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
122 * - Don't rerun crunchers that only depend on the request URL.
123 * - Don't count redirects and CGI requests as "blocked requests".
125 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
126 * Support crunching based on tags created by server-header taggers.
128 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
129 * Make HTTP snippets static (again). Add a Content-Type for those
130 * with content so the browser doesn't guess it based on the URL.
132 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
133 * Move crunching and request line rebuilding out of chat().
135 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
136 * Introduce tags as alternative way to specify which
137 * actions apply to a request. At the moment tags can be
138 * created based on client and server headers.
140 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
141 * Don't CLF-log CONNECT requests twice.
143 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
144 * Log the source address for ACL-related connection drops.
146 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
147 * New config option: enforce-blocks.
149 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
150 * - Move null byte check into separate function.
151 * - Don't confuse the client with error pages
152 * if a CONNECT request was already confirmed.
154 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
155 * Speed up NULL byte escaping and only log the complete
156 * NULL byte requests with header debugging enabled.
158 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
159 * Answer requests that contain NULL bytes with
160 * a custom response instead of waiting for more
161 * data until the client eventually hangs up.
163 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
164 * - Move delivery and logging of crunched responses
165 * from chat() into send_crunch_response().
166 * - Display the reason for generating http_responses.
167 * - Log the content length for LOG_LEVEL_CLF correctly
168 * (still incorrect for some fixed responses).
169 * - Reword an incorrect comment about
170 * treat-forbidden-connects-like-blocks violating
172 * - Add some log messages.
174 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
175 * Move mutex initialization into separate
176 * function and exit in case of errors.
178 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
179 * - Start to reduce chat()'s line count and move
180 * parts of it into separate functions.
181 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
183 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
184 * - Add Proxy-Agent header to HTTP snippets that are
185 * supposed to reach HTTP clients only.
186 * - Made a few CONNECT log messages more descriptive.
187 * - Catch completely empty server responses (as seen
188 * with Tor's fake ".noconnect" top level domain).
189 * - Use shiny new "forwarding-failed" template for socks errors.
191 * Revision 1.118 2007/01/07 07:43:43 joergs
192 * AmigaOS4 support added.
194 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
195 * Added config option accept-intercepted-requests
196 * and disabled it by default.
198 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
199 * Reverted parts of my last commit
200 * to keep error handling working.
202 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
203 * Fixed gcc43 conversion warnings.
205 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
206 * Fix -pedantic ISO C warning about converting
207 * from function pointer to object pointer.
209 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
210 * Silence compiler warning I introduced with my last commit.
212 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
213 * Mutex protect rand() if POSIX threading
214 * is used, warn the user if that's not possible
215 * and stop using it on _WIN32 where it could
218 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
219 * Don't prevent core dumps by catching SIGABRT.
220 * It's rude and makes debugging unreasonable painful.
222 * Revision 1.110 2006/12/13 14:52:53 etresoft
223 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
225 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
226 * Privoxy is now able to run as intercepting
227 * proxy in combination with any packet filter
228 * that does the port redirection. The destination
229 * is extracted from the "Host:" header which
230 * should be available for nearly all requests.
232 * Moved HTTP snipplets into jcc.c.
233 * Added error message for gopher proxy requests.
235 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
236 * Only unlink the pidfile if it's actually used.
238 * Change order of interception checks to make
239 * it possible to block or redirect requests for
242 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
243 * Make pthread mutex locking more generic. Instead of
244 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
245 * and use mutex locking unless there is an _r function
246 * available. Better safe than sorry.
248 * Fixes "./configure --disable-pthread" and should result
249 * in less threading-related problems on pthread-using platforms,
250 * but it still doesn't fix BR#1122404.
252 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
253 * Move pthread.h inclusion from jcc.c to jcc.h.
254 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
256 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
257 * Don't exit after receiving the second SIGHUP on Solaris.
259 * Fixes BR 1052235, but the same problem may exist on other
260 * systems. Once 3.0.6 is out we should use sigset()
261 * where available and see if it breaks anything.
263 * Revision 1.104 2006/09/23 13:26:38 roro
264 * Replace TABs by spaces in source code.
266 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
267 * Fix +redirect{}. Didn't work with -fast-redirects.
269 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
270 * Respond with 400 and a short text message
271 * if the client tries to use Privoxy as FTP proxy.
273 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
274 * Make number of retries in case of forwarded-connect problems
275 * a config file option (forwarded-connect-retries) and use 0 as
278 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
279 * Set random(3) seed.
281 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
282 * Follow the OpenBSD port's lead and protect the resolve
283 * functions on OpenBSD as well.
285 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
286 * --user fix. Only use the user as group if no group is specified.
287 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
289 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
290 * Windows service (re-)integration
294 * --install[:service_name]
295 * --uninstall[:service_name]
298 * They work as follows:
299 * --install will create a service for you and then terminate.
300 * By default the service name will be "privoxy" (without the quotes).
301 * However you can run multiple services if you wish, just by adding
302 * a colon and then a name (no spaces).
304 * --uninstall follows the exact same rules a --install.
306 * --service is used when the program is executed by the service
307 * control manager, and in normal circumstances would never be
308 * used as a command line argument.
310 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
311 * Windows service integration
313 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
314 * Incorporate Fabian Keil's patch work:
315 http://www.fabiankeil.de/sourcecode/privoxy/
317 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
318 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
319 * with what was really the latest development (the v_3_0_branch branch)
321 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
322 * Thanks to Jindrich Makovicka for a race condition fix for the log
323 * file. The race condition remains for non-pthread implementations.
324 * Reference patch #1175720.
326 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
327 * Add the ability to check jpeg images for invalid
328 * lengths of comment blocks. Defensive strategy
329 * against the exploit:
330 * Microsoft Security Bulletin MS04-028
331 * Buffer Overrun in JPEG Processing (GDI+) Could
332 * Allow Code Execution (833987)
333 * Enabled with +inspect-jpegs in actions files.
335 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
336 * - Fixed usage info for non-unix platforms
337 * - Fixed small cmdline parsing bug
339 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
340 * Diagnostics: Now preserve the returncode of pthread_create
341 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
343 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
344 * No longer ignore SIGCHLD. Fixes bug #769381
346 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
347 * Close jarfile on graceful exit, remove stray line
349 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
350 * Cosmetics: Killed a warning, a typo and an allocation left at exit
352 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
353 * No longer rely on non-POSIX.1 extensions of getcwd().
356 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
357 * Replaced setenv() by posix-compliant putenv()
358 * Thanks to Neil McCalden (nmcc AT users.sf.net).
360 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
361 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
363 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
364 * Clean-up and extension of improvements for forked mode:
365 * - Child's return code now consists of flags RC_FLAG_*
366 * - Reporting toggle to parent now properly #ifdef'ed
367 * - Children now report blocking to parent. This enables
368 * statistics in forked mode
370 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
371 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
372 * Children now report having been toggled through _exit(17), parents
373 * watch for that code and toggle themselves if found.
375 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
376 * Wrapping all *_r functions (the non-_r versions of them) with
377 * mutex semaphores for OSX. Hopefully this will take care of all
378 * of those pesky crash reports.
380 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
381 * Fixed two mostly harmless mem leaks
383 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
384 * Fix: Head of global clients list now initialized to NULL
386 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
387 * Added basic support for OPTIONS and TRACE HTTP methods:
388 * - New interceptor direct_response() added in chat().
389 * - sed() moved to earlier in the process, so that the
390 * Host: header is evaluated before actions and forwarding
393 * Revision 1.92 2002/05/08 16:00:46 oes
394 * Chat's buffer handling:
395 * - Fixed bug with unchecked out-of-mem conditions
396 * while reading client request & server headers
397 * - No longer predict if the buffer limit will be exceeded
398 * in the next read -- check add_to_iob's new
399 * return code. If buffer couldn't be extended
400 * (policy or out-of-mem) while
401 * - reading from client: abort
402 * - reading server headers: send error page
403 * - buffering server body for filter: flush,
404 * and if that fails: send error page
406 * Revision 1.91 2002/04/08 20:35:58 swa
409 * Revision 1.90 2002/04/02 14:57:28 oes
410 * Made sending wafers independent of FEATURE_COOKIE_JAR
412 * Revision 1.89 2002/03/31 17:18:59 jongfoster
413 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
415 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
416 * More compiler warning message maintenance
418 * Revision 1.87 2002/03/26 22:29:54 swa
419 * we have a new homepage!
421 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
422 * Workaround for closing the jarfile before load_config() comes around again
424 * Revision 1.85 2002/03/24 15:23:33 jongfoster
427 * Revision 1.84 2002/03/24 13:25:43 swa
428 * name change related issues
430 * Revision 1.83 2002/03/16 23:54:06 jongfoster
431 * Adding graceful termination feature, to help look for memory leaks.
432 * If you enable this (which, by design, has to be done by hand
433 * editing config.h) and then go to http://i.j.b/die, then the program
434 * will exit cleanly after the *next* request. It should free all the
435 * memory that was used.
437 * Revision 1.82 2002/03/13 00:27:05 jongfoster
440 * Revision 1.81 2002/03/12 01:42:50 oes
441 * Introduced modular filters
443 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
444 * OS/2 port maintenance:
445 * - Fixed EMX build - it had decayed a little
446 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
447 * substituted a memset for now.
449 * Revision 1.79 2002/03/09 20:03:52 jongfoster
450 * - Making various functions return int rather than size_t.
451 * (Undoing a recent change). Since size_t is unsigned on
452 * Windows, functions like read_socket that return -1 on
453 * error cannot return a size_t.
455 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
456 * crashes, and also frequently caused JB to jump to 100%
457 * CPU and stay there. (Because it thought it had just
458 * read ((unsigned)-1) == 4Gb of data...)
460 * - The signature of write_socket has changed, it now simply
461 * returns success=0/failure=nonzero.
463 * - Trying to get rid of a few warnings --with-debug on
464 * Windows, I've introduced a new type "jb_socket". This is
465 * used for the socket file descriptors. On Windows, this
466 * is SOCKET (a typedef for unsigned). Everywhere else, it's
467 * an int. The error value can't be -1 any more, so it's
468 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
469 * Windows it maps to the #define INVALID_SOCKET.)
471 * - The signature of bind_port has changed.
473 * Revision 1.78 2002/03/08 21:35:04 oes
474 * Added optional group supplement to --user option. Will now use default group of user if no group given
476 * Revision 1.77 2002/03/07 03:52:06 oes
477 * - Fixed compiler warnings etc
478 * - Improved handling of failed DNS lookups
480 * Revision 1.76 2002/03/06 22:54:35 jongfoster
481 * Automated function-comment nitpicking.
483 * Revision 1.75 2002/03/06 10:02:19 oes
484 * Fixed stupid bug when --user was not given
486 * Revision 1.74 2002/03/06 00:49:31 jongfoster
487 * Fixing warning on Windows
488 * Making #ifdefs that refer to the same variable consistently
489 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
491 * Revision 1.73 2002/03/05 23:57:30 hal9
492 * Stray character 's' on line 1618 was breaking build.
494 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
495 * - Re-enable OS/2 building after new parms were added
496 * - Fix false out of memory report when resolving CGI templates when no IP
497 * address is available of failed attempt (a la no such domain)
499 * Revision 1.71 2002/03/05 18:13:56 oes
500 * Added --user option
502 * Revision 1.70 2002/03/05 04:52:42 oes
503 * Deleted non-errlog debugging code
505 * Revision 1.69 2002/03/04 23:50:00 jongfoster
506 * Splitting off bind_port() call into bind_port_helper(), with
509 * Revision 1.68 2002/03/04 20:17:32 oes
512 * Revision 1.67 2002/03/04 18:18:57 oes
513 * - Removed _DEBUG mode
514 * - Cleand up cmdline parsing
515 * - Introduced --no-daemon, --pidfile options
516 * - Cleaned up signal handling:
517 * - Terminate cleanly on INT, TERM and ABRT
518 * - Schedule logfile for re-opening on HUP
519 * - Ignore CHLD and PIPE
520 * - Leave the rest with their default handlers
521 * - Uniform handler registration
522 * - Added usage() function
523 * - Played styleguide police
525 * Revision 1.66 2002/03/03 15:06:55 oes
526 * Re-enabled automatic config reloading
528 * Revision 1.65 2002/03/03 14:49:11 oes
529 * Fixed CLF logging: Now uses client's original HTTP request
531 * Revision 1.64 2002/03/03 09:18:03 joergs
532 * Made jumbjuster work on AmigaOS again.
534 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
535 * Clean up a little CRLF unpleasantness that suddenly appeared
537 * Revision 1.62 2002/02/20 23:17:23 jongfoster
538 * Detecting some out-of memory conditions and exiting with a log message.
540 * Revision 1.61 2002/01/17 21:01:52 jongfoster
541 * Moving all our URL and URL pattern parsing code to urlmatch.c.
543 * Revision 1.60 2001/12/30 14:07:32 steudten
544 * - Add signal handling (unix)
545 * - Add SIGHUP handler (unix)
546 * - Add creation of pidfile (unix)
547 * - Add action 'top' in rc file (RH)
548 * - Add entry 'SIGNALS' to manpage
549 * - Add exit message to logfile (unix)
551 * Revision 1.59 2001/12/13 14:07:18 oes
552 * Fixed Bug: 503 error page now sent OK
554 * Revision 1.58 2001/11/30 23:37:24 jongfoster
555 * Renaming the Win32 config file to config.txt - this is almost the
556 * same as the corresponding UNIX name "config"
558 * Revision 1.57 2001/11/16 00:47:43 jongfoster
559 * Changing the tty-disconnection code to use setsid().
561 * Revision 1.56 2001/11/13 20:20:54 jongfoster
562 * Tabs->spaces, fixing a bug with missing {} around an if()
564 * Revision 1.55 2001/11/13 20:14:53 jongfoster
565 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
567 * Revision 1.54 2001/11/07 00:03:14 steudten
568 * Give reliable return value if an error
569 * occurs not just 0 with new daemon mode.
571 * Revision 1.53 2001/11/05 21:41:43 steudten
572 * Add changes to be a real daemon just for unix os.
573 * (change cwd to /, detach from controlling tty, set
574 * process group and session leader to the own process.
576 * Add some fatal-error log message for failed malloc().
577 * Add '-d' if compiled with 'configure --with-debug' to
578 * enable debug output.
580 * Revision 1.52 2001/10/26 20:11:20 jongfoster
581 * Fixing type mismatch
583 * Revision 1.51 2001/10/26 17:38:28 oes
586 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
587 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
588 * threads to call select() simultaneously. So, it's time to do a real, live,
589 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
590 * (native). Both versions will work, but using __OS2__ offers multi-threading.
592 * Revision 1.49 2001/10/23 21:41:35 jongfoster
593 * Added call to initialize the (statically-allocated of course)
594 * "out of memory" CGI response.
596 * Revision 1.48 2001/10/10 19:56:46 jongfoster
597 * Moving some code that wasn't cookie-related out of an #ifdef
600 * Revision 1.47 2001/10/10 16:44:36 oes
601 * Added CONNECT destination port limitation check
603 * Revision 1.46 2001/10/08 15:17:41 oes
604 * Re-enabled SSL forwarding
606 * Revision 1.45 2001/10/07 15:42:11 oes
607 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
609 * Moved downgrading of the HTTP version from parse_http_request to
610 * chat(), since we can't decide if it is necessary before we have
611 * determined the actions for the URL. The HTTP command is now
612 * *always* re-built so the repairs need no longer be special-cased.
614 * filter_popups now gets a csp pointer so it can raise the new
615 * CSP_FLAG_MODIFIED flag.
619 * Added configurable size limit for the IOB. If the IOB grows so
620 * large that the next read would exceed the limit, the header
621 * is generated, and the header & unfiltered buffer are flushed
622 * to the client. Chat then continues in non-buffering,
623 * non-filtering body mode.
625 * Revision 1.44 2001/10/02 18:13:57 oes
628 * Revision 1.43 2001/10/02 15:32:13 oes
629 * Moved generation of hdr
631 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
632 * Cleaning up 2 compiler warnings on OS/2.
634 * Revision 1.41 2001/09/16 17:05:14 jongfoster
635 * Removing unused #include showarg.h
637 * Revision 1.40 2001/09/16 15:41:45 jongfoster
638 * Fixing signed/unsigned comparison warning.
640 * Revision 1.39 2001/09/16 13:21:27 jongfoster
641 * Changes to use new list functions.
643 * Revision 1.38 2001/09/16 13:01:46 jongfoster
644 * Removing redundant function call that zeroed zalloc()'d memory.
646 * Revision 1.37 2001/09/10 11:12:24 oes
647 * Deleted unused variable
649 * Revision 1.36 2001/09/10 10:56:15 oes
650 * Silenced compiler warnings
652 * Revision 1.35 2001/07/31 14:44:22 oes
653 * Deleted unused size parameter from filter_popups()
655 * Revision 1.34 2001/07/30 22:08:36 jongfoster
656 * Tidying up #defines:
657 * - All feature #defines are now of the form FEATURE_xxx
658 * - Permanently turned off WIN_GUI_EDIT
659 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
661 * Revision 1.33 2001/07/29 19:32:00 jongfoster
662 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
664 * Revision 1.32 2001/07/29 18:47:05 jongfoster
665 * Adding missing #include "loadcfg.h"
667 * Revision 1.31 2001/07/29 12:17:48 oes
668 * Applied pthread fix by Paul Lieverse
670 * Revision 1.30 2001/07/25 22:57:13 jongfoster
671 * __BEOS__ no longer overrides FEATURE_PTHREAD.
672 * This is because FEATURE_PTHREAD will soon be widely used, so I
673 * want to keep it simple.
675 * Revision 1.29 2001/07/24 12:47:06 oes
676 * Applied BeOS support update by Eugenia
678 * Revision 1.28 2001/07/23 13:26:12 oes
679 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
681 * Revision 1.27 2001/07/19 19:09:47 haroon
682 * - Added code to take care of the situation where while processing the first
683 * server response (which includes the server header), after finding the end
684 * of the headers we were not looking past the end of the headers for
685 * content modification. I enabled it for filter_popups.
686 * Someone else should look to see if other similar operations should be
687 * done to the discarded portion of the buffer.
689 * Note 2001/07/20: No, the other content modification mechanisms will process
690 * the whole iob later anyway. --oes
692 * Revision 1.26 2001/07/18 12:31:36 oes
695 * Revision 1.25 2001/07/15 19:43:49 jongfoster
696 * Supports POSIX threads.
697 * Also removed some unused #includes.
699 * Revision 1.24 2001/07/13 14:00:40 oes
700 * - Generic content modification scheme:
701 * Each feature has its own applicability flag that is set
702 * from csp->action->flags.
703 * Replaced the "filtering" int flag , by a function pointer
704 * "content_filter" to the function that will do the content
705 * modification. If it is != NULL, the document will be buffered
706 * and processed through *content_filter, which must set
707 * csp->content_length and return a modified copy of the body
708 * or return NULL (on failiure).
709 * - Changed csp->is_text to the more generic bitmap csp->content_type
710 * which can currently take the valued CT_TEXT or CT_GIF
712 * - Removed all #ifdef PCRS
714 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
715 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
716 * the `block_acl' call. This prevents a compilation error when the user
717 * does not wish to use the "ACL" feature.
719 * Revision 1.22 2001/06/29 21:45:41 oes
720 * Indentation, CRLF->LF, Tab-> Space
722 * Revision 1.21 2001/06/29 13:29:36 oes
723 * - Cleaned up, improved comments
724 * - Unified all possible interceptors (CGI,
725 * block, trust, fast_redirect) in one
726 * place, with one (CGI) answer generation
727 * mechansim. Much clearer now.
728 * - Removed the GIF image generation, which
729 * is now done in filters.c:block_url()
730 * - Made error conditions like domain lookup
731 * failiure or (various) problems while talking
732 * to the server use cgi.c:error_response()
733 * instead of generating HTML/HTTP in chat() (yuck!)
734 * - Removed logentry from cancelled commit
736 * Revision 1.20 2001/06/09 10:55:28 jongfoster
737 * Changing BUFSIZ ==> BUFFER_SIZE
739 * Revision 1.19 2001/06/07 23:12:52 jongfoster
740 * Replacing function pointer in struct gateway with a directly
741 * called function forwarded_connect().
742 * Replacing struct gateway with struct forward_spec
744 * Revision 1.18 2001/06/03 19:12:16 oes
745 * introduced new cgi handling
747 * Revision 1.17 2001/06/01 20:07:23 jongfoster
748 * Now uses action +image-blocker{} rather than config->tinygif
750 * Revision 1.16 2001/06/01 18:49:17 jongfoster
751 * Replaced "list_share" with "list" - the tiny memory gain was not
752 * worth the extra complexity.
754 * Revision 1.15 2001/05/31 21:24:47 jongfoster
755 * Changed "permission" to "action" throughout.
756 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
757 * Moved vanilla wafer check into chat(), since we must now
758 * decide whether or not to add it based on the URL.
760 * Revision 1.14 2001/05/29 20:14:01 joergs
761 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
764 * Revision 1.13 2001/05/29 09:50:24 jongfoster
765 * Unified blocklist/imagelist/permissionslist.
766 * File format is still under discussion, but the internal changes
769 * Also modified interceptor behaviour:
770 * - We now intercept all URLs beginning with one of the following
771 * prefixes (and *only* these prefixes):
773 * * http://ijbswa.sf.net/config/
774 * * http://ijbswa.sourceforge.net/config/
775 * - New interceptors "home page" - go to http://i.j.b/ to see it.
776 * - Internal changes so that intercepted and fast redirect pages
777 * are not replaced with an image.
778 * - Interceptors now have the option to send a binary page direct
779 * to the client. (i.e. ijb-send-banner uses this)
780 * - Implemented show-url-info interceptor. (Which is why I needed
781 * the above interceptors changes - a typical URL is
782 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
783 * The previous mechanism would not have intercepted that, and
784 * if it had been intercepted then it then it would have replaced
787 * Revision 1.12 2001/05/27 22:17:04 oes
789 * - re_process_buffer no longer writes the modified buffer
790 * to the client, which was very ugly. It now returns the
791 * buffer, which it is then written by chat.
793 * - content_length now adjusts the Content-Length: header
794 * for modified documents rather than crunch()ing it.
795 * (Length info in csp->content_length, which is 0 for
796 * unmodified documents)
798 * - For this to work, sed() is called twice when filtering.
800 * Revision 1.11 2001/05/26 17:27:53 jongfoster
801 * Added support for CLF and fixed LOG_LEVEL_LOG.
802 * Also did CRLF->LF fix of my previous patch.
804 * Revision 1.10 2001/05/26 15:26:15 jongfoster
805 * ACL feature now provides more security by immediately dropping
806 * connections from untrusted hosts.
808 * Revision 1.9 2001/05/26 00:28:36 jongfoster
809 * Automatic reloading of config file.
810 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
811 * Most of the global variables have been moved to a new
812 * struct configuration_spec, accessed through csp->config->globalname
813 * Most of the globals remaining are used by the Win32 GUI.
815 * Revision 1.8 2001/05/25 22:43:18 jongfoster
816 * Fixing minor memory leak and buffer overflow.
818 * Revision 1.7 2001/05/25 22:34:30 jongfoster
821 * Revision 1.6 2001/05/23 00:13:58 joergs
822 * AmigaOS support fixed.
824 * Revision 1.5 2001/05/22 18:46:04 oes
826 * - Enabled filtering banners by size rather than URL
827 * by adding patterns that replace all standard banner
828 * sizes with the "Junkbuster" gif to the re_filterfile
830 * - Enabled filtering WebBugs by providing a pattern
831 * which kills all 1x1 images
833 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
834 * which is selected by the (nonstandard and therefore
835 * capital) letter 'U' in the option string.
836 * It causes the quantifiers to be ungreedy by default.
837 * Appending a ? turns back to greedy (!).
839 * - Added a new interceptor ijb-send-banner, which
840 * sends back the "Junkbuster" gif. Without imagelist or
841 * MSIE detection support, or if tinygif = 1, or the
842 * URL isn't recognized as an imageurl, a lame HTML
843 * explanation is sent instead.
845 * - Added new feature, which permits blocking remote
846 * script redirects and firing back a local redirect
848 * The feature is conditionally compiled, i.e. it
849 * can be disabled with --disable-fast-redirects,
850 * plus it must be activated by a "fast-redirects"
851 * line in the config file, has its own log level
852 * and of course wants to be displayed by show-proxy-args
853 * Note: Boy, all the #ifdefs in 1001 locations and
854 * all the fumbling with configure.in and acconfig.h
855 * were *way* more work than the feature itself :-(
857 * - Because a generic redirect template was needed for
858 * this, tinygif = 3 now uses the same.
860 * - Moved GIFs, and other static HTTP response templates
865 * - Removed some >400 CRs again (Jon, you really worked
868 * Revision 1.4 2001/05/21 19:34:01 jongfoster
869 * Made failure to bind() a fatal error.
871 * Revision 1.3 2001/05/20 01:21:20 jongfoster
872 * Version 2.9.4 checkin.
873 * - Merged popupfile and cookiefile, and added control over PCRS
874 * filtering, in new "permissionsfile".
875 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
876 * file error you now get a message box (in the Win32 GUI) rather
877 * than the program exiting with no explanation.
878 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
880 * - Removed tabs from "config"
881 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
882 * - Bumped up version number.
884 * Revision 1.2 2001/05/17 22:34:44 oes
885 * - Added hint on GIF char array generation to jcc.c
886 * - Cleaned CRLF's from the sources and related files
887 * - Repaired logging for REF and FRC
889 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
890 * Initial import of version 2.9.3 source tree
893 *********************************************************************/
899 #include <sys/types.h>
908 # ifndef FEATURE_PTHREAD
912 # include <windows.h>
913 # include <process.h>
914 # endif /* ndef FEATURE_PTHREAD */
917 # ifndef _WIN_CONSOLE
919 # endif /* ndef _WIN_CONSOLE */
920 # include "w32svrapi.h"
922 #else /* ifndef _WIN32 */
924 # if !defined (__OS2__)
926 # include <sys/wait.h>
927 # endif /* ndef __OS2__ */
928 # include <sys/time.h>
929 # include <sys/stat.h>
930 # include <sys/ioctl.h>
933 #include <sys/termios.h>
944 # include <socket.h> /* BeOS has select() for sockets only. */
945 # include <OS.h> /* declarations for threads and stuff. */
948 # if defined(__EMX__) || defined(__OS2__)
949 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
954 #define bzero(B,N) memset(B,0x00,n)
969 #include "killpopup.h"
970 #include "miscutil.h"
972 #include "jbsockets.h"
977 #include "urlmatch.h"
979 const char jcc_h_rcs[] = JCC_H_VERSION;
980 const char project_h_rcs[] = PROJECT_H_VERSION;
983 struct client_state clients[1];
984 struct file_list files[1];
986 #ifdef FEATURE_STATISTICS
987 int urls_read = 0; /* total nr of urls read inc rejected */
988 int urls_rejected = 0; /* total nr of urls rejected */
989 #endif /* def FEATURE_STATISTICS */
991 #ifdef FEATURE_GRACEFUL_TERMINATION
995 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
996 static void sig_handler(int the_signal);
998 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
999 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1000 static jb_err get_server_headers(struct client_state *csp);
1001 static const char *crunch_reason(const struct http_response *rsp);
1002 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
1004 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1006 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1007 static jb_err change_request_destination(struct client_state *csp);
1008 static void chat(struct client_state *csp);
1009 static void serve(struct client_state *csp);
1011 static void usage(const char *myname);
1013 static void initialize_mutexes(void);
1014 static jb_socket bind_port_helper(struct configuration_spec *config);
1015 static void listen_loop(void);
1018 void serve(struct client_state *csp);
1019 #else /* ifndef AMIGA */
1020 static void serve(struct client_state *csp);
1021 #endif /* def AMIGA */
1024 static int32 server_thread(void *data);
1025 #endif /* def __BEOS__ */
1028 #define sleep(N) Sleep(((N) * 1000))
1032 #define sleep(N) DosSleep(((N) * 100))
1035 #ifdef FEATURE_PTHREAD
1036 pthread_mutex_t log_mutex;
1037 pthread_mutex_t log_init_mutex;
1039 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1040 pthread_mutex_t resolver_mutex;
1041 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1043 #ifndef HAVE_GMTIME_R
1044 pthread_mutex_t gmtime_mutex;
1045 #endif /* ndef HAVE_GMTIME_R */
1047 #ifndef HAVE_LOCALTIME_R
1048 pthread_mutex_t localtime_mutex;
1049 #endif /* ndef HAVE_GMTIME_R */
1052 pthread_mutex_t rand_mutex;
1053 #endif /* ndef HAVE_RANDOM */
1055 #endif /* FEATURE_PTHREAD */
1057 #if defined(unix) || defined(__EMX__)
1058 const char *basedir = NULL;
1059 const char *pidfile = NULL;
1060 int received_hup_signal = 0;
1061 #endif /* defined unix */
1063 /* The vanilla wafer. */
1064 static const char VANILLA_WAFER[] =
1065 "NOTICE=TO_WHOM_IT_MAY_CONCERN_"
1066 "Do_not_send_me_any_copyrighted_information_other_than_the_"
1067 "document_that_I_am_requesting_or_any_of_its_necessary_components._"
1068 "In_particular_do_not_send_me_any_cookies_that_"
1069 "are_subject_to_a_claim_of_copyright_by_anybody._"
1070 "Take_notice_that_I_refuse_to_be_bound_by_any_license_condition_"
1071 "(copyright_or_otherwise)_applying_to_any_cookie._";
1073 /* HTTP snipplets. */
1074 static const char CSUCCEED[] =
1075 "HTTP/1.0 200 Connection established\n"
1076 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1078 static const char CHEADER[] =
1079 "HTTP/1.0 400 Invalid header received from browser\r\n"
1080 "Proxy-Agent: Privoxy " VERSION "\r\n"
1081 "Content-Type: text/plain\r\n"
1082 "Connection: close\r\n\r\n"
1083 "Invalid header received from browser.\r\n";
1085 static const char CFORBIDDEN[] =
1086 "HTTP/1.0 403 Connection not allowable\r\n"
1087 "Proxy-Agent: Privoxy " VERSION "\r\n"
1088 "X-Hint: If you read this message interactively, then you know why this happens ,-)\r\n"
1089 "Connection: close\r\n\r\n";
1091 static const char FTP_RESPONSE[] =
1092 "HTTP/1.0 400 Invalid request received from browser\r\n"
1093 "Content-Type: text/plain\r\n"
1094 "Connection: close\r\n\r\n"
1095 "Invalid request. Privoxy doesn't support FTP.\r\n";
1097 static const char GOPHER_RESPONSE[] =
1098 "HTTP/1.0 400 Invalid request received from browser\r\n"
1099 "Content-Type: text/plain\r\n"
1100 "Connection: close\r\n\r\n"
1101 "Invalid request. Privoxy doesn't support gopher.\r\n";
1103 /* XXX: should be a template */
1104 static const char MISSING_DESTINATION_RESPONSE[] =
1105 "HTTP/1.0 400 Bad request received from browser\r\n"
1106 "Proxy-Agent: Privoxy " VERSION "\r\n"
1107 "Content-Type: text/plain\r\n"
1108 "Connection: close\r\n\r\n"
1109 "Bad request. Privoxy was unable to extract the destination.\r\n";
1111 /* XXX: should be a template */
1112 static const char NO_SERVER_DATA_RESPONSE[] =
1113 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1114 "Proxy-Agent: Privoxy " VERSION "\r\n"
1115 "Content-Type: text/plain\r\n"
1116 "Connection: close\r\n\r\n"
1117 "Empty server or forwarder response.\r\n"
1118 "The connection was closed without sending any data.\r\n";
1121 /* XXX: should be a template */
1122 static const char NULL_BYTE_RESPONSE[] =
1123 "HTTP/1.0 400 Bad request received from browser\r\n"
1124 "Proxy-Agent: Privoxy " VERSION "\r\n"
1125 "Content-Type: text/plain\r\n"
1126 "Connection: close\r\n\r\n"
1127 "Bad request. Null byte(s) before end of request.\r\n";
1130 /* XXX: should be a template */
1131 static const char MESSED_UP_REQUEST_RESPONSE[] =
1132 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1133 "Proxy-Agent: Privoxy " VERSION "\r\n"
1134 "Content-Type: text/plain\r\n"
1135 "Connection: close\r\n\r\n"
1136 "Bad request. Messed up with header filters.\r\n";
1138 /* A function to crunch a response */
1139 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1141 /* Crunch function flags */
1142 #define CF_NO_FLAGS 0
1143 /* Cruncher applies to forced requests as well */
1144 #define CF_IGNORE_FORCE 1
1145 /* Crunched requests are counted for the block statistics */
1146 #define CF_COUNT_AS_REJECT 2
1148 /* A crunch function and its flags */
1151 const crunch_func_ptr cruncher;
1155 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1157 /* Complete list of cruncher functions */
1158 static const struct cruncher crunchers_all[] = {
1159 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1160 { block_url, CF_COUNT_AS_REJECT },
1161 #ifdef FEATURE_TRUST
1162 { trust_url, CF_COUNT_AS_REJECT },
1163 #endif /* def FEATURE_TRUST */
1164 { redirect_url, CF_NO_FLAGS },
1165 { dispatch_cgi, CF_IGNORE_FORCE},
1169 /* Light version, used after tags are applied */
1170 static const struct cruncher crunchers_light[] = {
1171 { block_url, CF_COUNT_AS_REJECT },
1172 { redirect_url, CF_NO_FLAGS },
1177 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1178 /*********************************************************************
1180 * Function : sig_handler
1182 * Description : Signal handler for different signals.
1183 * Exit gracefully on TERM and INT
1184 * or set a flag that will cause the errlog
1185 * to be reopened by the main thread on HUP.
1188 * 1 : the_signal = the signal cause this function to call
1192 *********************************************************************/
1193 static void sig_handler(int the_signal)
1199 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1210 received_hup_signal = 1;
1215 * We shouldn't be here, unless we catch signals
1216 * in main() that we can't handle here!
1218 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1226 /*********************************************************************
1228 * Function : client_protocol_is_unsupported
1230 * Description : Checks if the client used a known unsupported
1231 * protocol and deals with it by sending an error
1235 * 1 : csp = Current client state (buffers, headers, etc...)
1236 * 2 : req = the first request line send by the client
1238 * Returns : TRUE if an error response has been generated, or
1239 * FALSE if the request doesn't look invalid.
1241 *********************************************************************/
1242 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1244 char buf[BUFFER_SIZE];
1247 * If it's a FTP or gopher request, we don't support it.
1249 * These checks are better than nothing, but they might
1250 * not work in all configurations and some clients might
1251 * have problems digesting the answer.
1253 * They should, however, never cause more problems than
1254 * Privoxy's old behaviour (returning the misleading HTML
1257 * "Could not resolve http://(ftp|gopher)://example.org").
1259 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1261 if (!strncmpic(req, "GET ftp://", 10))
1263 strlcpy(buf, FTP_RESPONSE, sizeof(buf));
1264 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1265 csp->ip_addr_str, req);
1269 strlcpy(buf, GOPHER_RESPONSE, sizeof(buf));
1270 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1271 csp->ip_addr_str, req);
1273 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1275 write_socket(csp->cfd, buf, strlen(buf));
1284 /*********************************************************************
1286 * Function : get_request_destination_elsewhere
1288 * Description : If the client's request was redirected into
1289 * Privoxy without the client's knowledge,
1290 * the request line lacks the destination host.
1292 * This function tries to get it elsewhere,
1293 * provided accept-intercepted-requests is enabled.
1295 * "Elsewhere" currently only means "Host: header",
1296 * but in the future we may ask the redirecting
1297 * packet filter to look the destination up.
1299 * If the destination stays unknown, an error
1300 * response is send to the client and headers
1301 * are freed so that chat() can return directly.
1304 * 1 : csp = Current client state (buffers, headers, etc...)
1305 * 2 : headers = a header list
1307 * Returns : JB_ERR_OK if the destination is now known, or
1308 * JB_ERR_PARSE if it isn't.
1310 *********************************************************************/
1311 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1315 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1317 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1318 " Privoxy isn't configured to accept intercepted requests.",
1319 csp->ip_addr_str, csp->http->cmd);
1320 /* XXX: Use correct size */
1321 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1322 csp->ip_addr_str, csp->http->cmd);
1324 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1325 destroy_list(headers);
1327 return JB_ERR_PARSE;
1329 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1331 /* Split the domain we just got for pattern matching */
1332 init_domain_components(csp->http);
1338 /* We can't work without destination. Go spread the news.*/
1340 req = list_to_text(headers);
1342 /* XXX: Use correct size */
1343 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1344 csp->ip_addr_str, csp->http->cmd);
1345 log_error(LOG_LEVEL_ERROR,
1346 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1347 csp->ip_addr_str, csp->http->cmd, req);
1350 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1351 destroy_list(headers);
1353 return JB_ERR_PARSE;
1356 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1357 * to get the destination IP address, use it as host directly
1358 * or do a reverse DNS lookup first.
1363 /*********************************************************************
1365 * Function : get_server_headers
1367 * Description : Parses server headers in iob and fills them
1368 * into csp->headers so that they can later be
1372 * 1 : csp = Current client state (buffers, headers, etc...)
1374 * Returns : JB_ERR_OK if everything went fine, or
1375 * JB_ERR_PARSE if the headers were incomplete.
1377 *********************************************************************/
1378 static jb_err get_server_headers(struct client_state *csp)
1380 int continue_hack_in_da_house = 0;
1383 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1388 * continue hack in da house. Ignore the ending of
1389 * this head and continue enlisting header lines.
1390 * The reason is described below.
1392 enlist(csp->headers, "");
1393 continue_hack_in_da_house = 0;
1396 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1399 * It's a bodyless continue response, don't
1400 * stop header parsing after reaching it's end.
1402 * As a result Privoxy will concatenate the
1403 * next response's head and parse and deliver
1404 * the headers as if they belonged to one request.
1406 * The client will separate them because of the
1407 * empty line between them.
1409 * XXX: What we're doing here is clearly against
1410 * the intended purpose of the continue header,
1411 * and under some conditions (HTTP/1.0 client request)
1412 * it's a standard violation.
1414 * Anyway, "sort of against the spec" is preferable
1415 * to "always getting confused by Continue responses"
1416 * (Privoxy's behaviour before this hack was added)
1418 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1419 continue_hack_in_da_house = 1;
1421 else if (*header == '\0')
1424 * If the header is empty, but the Continue hack
1425 * isn't active, we can assume that we reached the
1426 * end of the buffer before we hit the end of the
1429 * Inform the caller an let it decide how to handle it.
1431 return JB_ERR_PARSE;
1435 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1438 * XXX: Should we quit the request and return a
1439 * out of memory error page instead?
1441 log_error(LOG_LEVEL_ERROR,
1442 "Out of memory while enlisting server headers. %s lost.",
1452 /*********************************************************************
1454 * Function : crunch_reason
1456 * Description : Translates the crunch reason code into a string.
1459 * 1 : rsp = a http_response
1461 * Returns : A string with the crunch reason or an error description.
1463 *********************************************************************/
1464 static const char *crunch_reason(const struct http_response *rsp)
1466 char * reason = NULL;
1468 assert(rsp != NULL);
1471 return "Internal error while searching for crunch reason";
1474 switch (rsp->reason)
1476 case RSP_REASON_UNSUPPORTED:
1477 reason = "Unsupported HTTP feature";
1479 case RSP_REASON_BLOCKED:
1482 case RSP_REASON_UNTRUSTED:
1483 reason = "Untrusted";
1485 case RSP_REASON_REDIRECTED:
1486 reason = "Redirected";
1488 case RSP_REASON_CGI_CALL:
1489 reason = "CGI Call";
1491 case RSP_REASON_NO_SUCH_DOMAIN:
1492 reason = "DNS failure";
1494 case RSP_REASON_FORWARDING_FAILED:
1495 reason = "Forwarding failed";
1497 case RSP_REASON_CONNECT_FAILED:
1498 reason = "Connection failure";
1500 case RSP_REASON_OUT_OF_MEMORY:
1501 reason = "Out of memory (may mask other reasons)";
1504 reason = "No reason recorded";
1512 /*********************************************************************
1514 * Function : send_crunch_response
1516 * Description : Delivers already prepared response for
1517 * intercepted requests, logs the interception
1518 * and frees the response.
1521 * 1 : csp = Current client state (buffers, headers, etc...)
1522 * 1 : rsp = Fully prepared response. Will be freed on exit.
1524 * Returns : Nothing.
1526 *********************************************************************/
1527 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1529 const struct http_request *http = csp->http;
1530 char status_code[4];
1532 assert(rsp != NULL);
1533 assert(rsp->head != NULL);
1538 * Not supposed to happen. If it does
1539 * anyway, treat it as an unknown error.
1541 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1542 /* return code doesn't matter */
1547 /* If rsp is still NULL, we have serious internal problems. */
1548 log_error(LOG_LEVEL_FATAL,
1549 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1553 * Extract the status code from the actual head
1554 * that was send to the client. It is the only
1555 * way to get it right for all requests, including
1556 * the fixed ones for out-of-memory problems.
1558 * A head starts like this: 'HTTP/1.1 200...'
1562 status_code[0] = rsp->head[9];
1563 status_code[1] = rsp->head[10];
1564 status_code[2] = rsp->head[11];
1565 status_code[3] = '\0';
1567 /* Write the answer to the client */
1568 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1569 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1571 /* There is nothing we can do about it. */
1572 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1575 /* Log that the request was crunched and why. */
1576 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1577 http->hostport, http->path, crunch_reason(rsp));
1578 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1579 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1581 /* Clean up and return */
1582 if (cgi_error_memory() != rsp)
1584 free_http_response(rsp);
1591 /*********************************************************************
1593 * Function : request_contains_null_bytes
1595 * Description : Checks for NULL bytes in the request and sends
1596 * an error message to the client if any were found.
1598 * XXX: currently not used, see comment in chat().
1601 * 1 : csp = Current client state (buffers, headers, etc...)
1602 * 2 : buf = Data from the client's request to check.
1603 * 3 : len = The data length.
1605 * Returns : TRUE if the request contained one or more NULL bytes, or
1608 *********************************************************************/
1609 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1611 size_t c_len; /* Request lenght when treated as C string */
1613 c_len = strlen(buf);
1618 * Null byte(s) found. Log the request,
1619 * return an error response and hang up.
1621 size_t tmp_len = c_len;
1626 * Replace NULL byte(s) with '°' characters
1627 * so the request can be logged as string.
1628 * XXX: Is there a better replacement character?
1631 tmp_len += strlen(buf+tmp_len);
1632 } while (tmp_len < len);
1634 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1635 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1636 log_error(LOG_LEVEL_HEADER,
1637 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1639 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1641 /* XXX: Log correct size */
1642 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1652 /*********************************************************************
1654 * Function : crunch_response_triggered
1656 * Description : Checks if the request has to be crunched,
1657 * and delivers the crunch response if necessary.
1660 * 1 : csp = Current client state (buffers, headers, etc...)
1661 * 2 : crunchers = list of cruncher functions to run
1663 * Returns : TRUE if the request was answered with a crunch response
1666 *********************************************************************/
1667 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1669 struct http_response *rsp = NULL;
1670 const struct cruncher *c;
1673 * If CGI request crunching is disabled,
1674 * check the CGI dispatcher out of order to
1675 * prevent unintentional blocks or redirects.
1677 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1678 && (NULL != (rsp = dispatch_cgi(csp))))
1680 /* Deliver, log and free the interception response. */
1681 send_crunch_response(csp, rsp);
1685 for (c = crunchers; c->cruncher != NULL; c++)
1688 * Check the cruncher if either Privoxy is toggled
1689 * on and the request isn't forced, or if the cruncher
1690 * applies to forced requests as well.
1692 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1693 !(csp->flags & CSP_FLAG_FORCED)) ||
1694 (c->flags & CF_IGNORE_FORCE))
1696 rsp = c->cruncher(csp);
1699 /* Deliver, log and free the interception response. */
1700 send_crunch_response(csp, rsp);
1701 #ifdef FEATURE_STATISTICS
1702 if (c->flags & CF_COUNT_AS_REJECT)
1704 csp->flags |= CSP_FLAG_REJECTED;
1706 #endif /* def FEATURE_STATISTICS */
1717 /*********************************************************************
1719 * Function : build_request_line
1721 * Description : Builds the HTTP request line.
1723 * If a HTTP forwarder is used it expects the whole URL,
1724 * web servers only get the path.
1727 * 1 : csp = Current client state (buffers, headers, etc...)
1728 * 2 : fwd = The forwarding spec used for the request
1729 * XXX: Should use http->fwd instead.
1730 * 3 : request_line = The old request line which will be replaced.
1732 * Returns : Nothing. Terminates in case of memory problems.
1734 *********************************************************************/
1735 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1737 struct http_request *http = csp->http;
1739 assert(http->ssl == 0);
1742 * Downgrade http version from 1.1 to 1.0
1743 * if +downgrade action applies.
1745 if ( (csp->action->flags & ACTION_DOWNGRADE)
1746 && (!strcmpic(http->ver, "HTTP/1.1")))
1749 http->ver = strdup("HTTP/1.0");
1751 if (http->ver == NULL)
1753 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1758 * Rebuild the request line.
1760 freez(*request_line);
1761 *request_line = strdup(http->gpc);
1762 string_append(request_line, " ");
1764 if (fwd->forward_host)
1766 string_append(request_line, http->url);
1770 string_append(request_line, http->path);
1772 string_append(request_line, " ");
1773 string_append(request_line, http->ver);
1775 if (*request_line == NULL)
1777 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1779 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1783 /*********************************************************************
1785 * Function : change_request_destination
1787 * Description : Parse a (rewritten) request line and regenerate
1788 * the http request data.
1791 * 1 : csp = Current client state (buffers, headers, etc...)
1793 * Returns : Forwards the parse_http_request() return code.
1794 * Terminates in case of memory problems.
1796 *********************************************************************/
1797 static jb_err change_request_destination(struct client_state *csp)
1799 struct http_request *http = csp->http;
1802 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1803 free_http_request(http);
1804 err = parse_http_request(csp->headers->first->str, http, csp);
1805 if (JB_ERR_OK != err)
1807 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1808 jb_err_to_string(err));
1810 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1811 if (http->ocmd == NULL)
1813 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1820 /*********************************************************************
1824 * Description : Once a connection to the client has been accepted,
1825 * this function is called (via serve()) to handle the
1826 * main business of the communication. When this
1827 * function returns, the caller must close the client
1830 * FIXME: chat is nearly thousand lines long.
1834 * 1 : csp = Current client state (buffers, headers, etc...)
1836 * Returns : Nothing.
1838 *********************************************************************/
1839 static void chat(struct client_state *csp)
1841 char buf[BUFFER_SIZE];
1849 int ms_iis5_hack = 0;
1850 size_t byte_count = 0;
1851 int forwarded_connect_retries = 0;
1852 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1853 const struct forward_spec * fwd;
1854 struct http_request *http;
1855 int len; /* for buffer sizes (and negative error codes) */
1857 #ifdef FEATURE_KILL_POPUPS
1858 int block_popups_now = 0; /* bool, 1==currently blocking popups */
1859 #endif /* def FEATURE_KILL_POPUPS */
1861 /* Function that does the content filtering for the current request */
1862 filter_function_ptr content_filter = NULL;
1864 /* Skeleton for HTTP response, if we should intercept the request */
1865 struct http_response *rsp;
1867 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1868 struct list header_list;
1869 struct list *headers = &header_list;
1873 memset(buf, 0, sizeof(buf));
1876 * Read the client's request. Note that since we're not using select() we
1877 * could get blocked here if a client connected, then didn't say anything!
1882 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1884 if (len <= 0) break; /* error! */
1887 * If there is no memory left for buffering the
1888 * request, there is nothing we can do but hang up
1890 if (add_to_iob(csp, buf, len))
1895 req = get_header(csp);
1897 } while ((NULL != req) && ('\0' == *req));
1901 /* Request received. Validate and parse it. */
1905 * XXX: Temporary disabled to prevent problems
1906 * with POST requests whose bodies are allowed to
1907 * contain NULL bytes. BR#1730105.
1909 * The main purpose of this check is to properly
1910 * log stuff like BitTorrent traffic and other junk
1911 * that hits public proxies. It's not required for
1912 * Privoxy to functions as those requests are discarded
1915 * It probably should be rewritten to only check
1916 * the head of the request. Another option would
1917 * be to let all POST requests pass, although that
1918 * may not be good enough.
1920 if (request_contains_null_bytes(csp, buf, len))
1922 /* NULL bytes found and dealt with, just hang up. */
1927 /* Does the request line look invalid? */
1928 if (client_protocol_is_unsupported(csp, req))
1931 * Yes. The request has already been
1932 * answered with a error response, the buffers
1933 * were freed and we're done with chatting.
1938 #ifdef FEATURE_FORCE_LOAD
1940 * If this request contains the FORCE_PREFIX and blocks
1941 * aren't enforced, get rid of it and set the force flag.
1943 if (strstr(req, FORCE_PREFIX))
1945 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1947 log_error(LOG_LEVEL_FORCE,
1948 "Ignored force prefix in request: \"%s\".", req);
1952 strclean(req, FORCE_PREFIX);
1953 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
1954 csp->flags |= CSP_FLAG_FORCED;
1958 #endif /* def FEATURE_FORCE_LOAD */
1959 err = parse_http_request(req, http, csp);
1960 if (JB_ERR_OK != err)
1962 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
1968 if (http->cmd == NULL)
1970 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1971 /* XXX: Use correct size */
1972 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1973 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
1975 free_http_request(http);
1979 /* grab the rest of the client's headers */
1983 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
1985 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1988 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
1993 * If there is no memory left for buffering the
1994 * request, there is nothing we can do but hang up
1996 if (add_to_iob(csp, buf, len))
2003 if (p == NULL) break;
2010 if (http->host == NULL)
2013 * If we still don't know the request destination,
2014 * the request is invalid or the client uses
2015 * Privoxy without its knowledge.
2017 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2020 * Our attempts to get the request destination
2021 * elsewhere failed or Privoxy is configured
2022 * to only accept proxy requests.
2024 * An error response has already been send
2025 * and we're done here.
2032 * Determine the actions for this URL
2034 #ifdef FEATURE_TOGGLE
2035 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2037 /* Most compatible set of actions (i.e. none) */
2038 init_current_action(csp->action);
2041 #endif /* ndef FEATURE_TOGGLE */
2043 url_actions(http, csp);
2047 * Save a copy of the original request for logging
2049 http->ocmd = strdup(http->cmd);
2051 if (http->ocmd == NULL)
2053 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2056 enlist(csp->headers, http->cmd);
2058 /* Append the previously read headers */
2059 list_append_list_unique(csp->headers, headers);
2060 destroy_list(headers);
2063 * If the user has not supplied any wafers, and the user has not
2064 * told us to suppress the vanilla wafer, then send the vanilla wafer.
2066 if (list_is_empty(csp->action->multi[ACTION_MULTI_WAFER])
2067 && ((csp->action->flags & ACTION_VANILLA_WAFER) != 0))
2069 enlist(csp->action->multi[ACTION_MULTI_WAFER], VANILLA_WAFER);
2072 err = sed(client_patterns, add_client_headers, csp);
2073 if (JB_ERR_OK != err)
2075 assert(err == JB_ERR_PARSE);
2076 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2078 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2080 /* Check request line for rewrites. */
2081 if ((NULL == csp->headers->first->str)
2082 || (strcmp(http->cmd, csp->headers->first->str) &&
2083 (JB_ERR_OK != change_request_destination(csp))))
2086 * A header filter broke the request line - bail out.
2088 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2089 /* XXX: Use correct size */
2090 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2091 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2093 free_http_request(http);
2097 /* decide how to route the HTTP request */
2098 if (NULL == (fwd = forward_url(http, csp)))
2100 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2101 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2105 * build the http request to send to the server
2106 * we have to do one of the following:
2108 * create = use the original HTTP request to create a new
2109 * HTTP request that has either the path component
2110 * without the http://domainspec (w/path) or the
2111 * full orininal URL (w/url)
2112 * Note that the path and/or the HTTP version may
2113 * have been altered by now.
2115 * connect = Open a socket to the host:port of the server
2116 * and short-circuit server and client socket.
2118 * pass = Pass the request unchanged if forwarding a CONNECT
2119 * request to a parent proxy. Note that we'll be sending
2120 * the CFAIL message ourselves if connecting to the parent
2121 * fails, but we won't send a CSUCCEED message if it works,
2122 * since that would result in a double message (ours and the
2123 * parent's). After sending the request to the parent, we simply
2126 * here's the matrix:
2129 * +--------+--------+
2131 * 0 | create | connect|
2133 * Forwarding +--------+--------+
2135 * 1 | create | pass |
2137 * +--------+--------+
2142 * Check if a CONNECT request is allowable:
2143 * In the absence of a +limit-connect action, allow only port 443.
2144 * If there is an action, allow whatever matches the specificaton.
2148 if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443)
2149 || (csp->action->flags & ACTION_LIMIT_CONNECT
2150 && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) )
2152 if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS)
2155 * The response may confuse some clients,
2156 * but makes unblocking easier.
2158 * XXX: It seems to work with all major browsers,
2159 * so we should consider returning a body by default someday ...
2161 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2162 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2163 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2165 csp->action->flags |= ACTION_BLOCK;
2170 write_socket(csp->cfd, CFORBIDDEN, strlen(CFORBIDDEN));
2171 log_error(LOG_LEVEL_INFO, "Request from %s denied. "
2172 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2173 csp->ip_addr_str, csp->action->string[ACTION_STRING_LIMIT_CONNECT],
2175 assert(NULL != csp->http->ocmd);
2176 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 403 0", csp->ip_addr_str, csp->http->ocmd);
2178 list_remove_all(csp->headers);
2180 * XXX: For consistency we might want to log a crunch message here.
2189 freez(csp->headers->first->str);
2190 build_request_line(csp, fwd, &csp->headers->first->str);
2193 hdr = list_to_text(csp->headers);
2196 /* FIXME Should handle error properly */
2197 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2201 * We have a request. Check if one of the crunchers wants it.
2203 if (crunch_response_triggered(csp, crunchers_all))
2206 * Yes. The client got the crunch response
2207 * and we are done here after cleaning up.
2210 list_remove_all(csp->headers);
2216 * The headers can't be removed earlier because
2217 * they were still needed for the referrer check
2218 * in case of CGI crunches.
2220 * XXX: Would it be worth to move the referrer check
2221 * into client_referrer() and set a flag if it's trusted?
2223 list_remove_all(csp->headers);
2225 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2227 if (fwd->forward_host)
2229 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2230 fwd->forward_host, fwd->forward_port, http->hostport);
2234 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2237 /* here we connect to the server, gateway, or the forwarder */
2239 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2240 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2242 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2243 forwarded_connect_retries, http->hostport);
2246 if (csp->sfd == JB_INVALID_SOCKET)
2248 if (fwd->type != SOCKS_NONE)
2251 rsp = error_response(csp, "forwarding-failed", errno);
2253 else if (errno == EINVAL)
2255 rsp = error_response(csp, "no-such-domain", errno);
2259 rsp = error_response(csp, "connect-failed", errno);
2260 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2265 /* Write the answer to the client */
2268 send_crunch_response(csp, rsp);
2275 if (fwd->forward_host || (http->ssl == 0))
2277 /* write the client's (modified) header to the server
2278 * (along with anything else that may be in the buffer)
2281 if (write_socket(csp->sfd, hdr, strlen(hdr))
2282 || (flush_socket(csp->sfd, csp) < 0))
2284 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2287 rsp = error_response(csp, "connect-failed", errno);
2291 send_crunch_response(csp, rsp);
2301 * We're running an SSL tunnel and we're not forwarding,
2302 * so just send the "connect succeeded" message to the
2303 * client, flush the rest, and get out of the way.
2305 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2313 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2315 /* we're finished with the client's header */
2318 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2320 /* pass data between the client and server
2321 * until one or the other shuts down the connection.
2330 * FD_ZERO here seems to point to an errant macro which crashes.
2331 * So do this by hand for now...
2333 memset(&rfds,0x00,sizeof(fd_set));
2337 FD_SET(csp->cfd, &rfds);
2338 FD_SET(csp->sfd, &rfds);
2340 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2344 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2348 /* this is the body of the browser's request
2349 * just read it and write it.
2352 if (FD_ISSET(csp->cfd, &rfds))
2354 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2358 break; /* "game over, man" */
2361 if (write_socket(csp->sfd, buf, (size_t)len))
2363 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2370 * The server wants to talk. It could be the header or the body.
2371 * If `hdr' is null, then it's the header otherwise it's the body.
2372 * FIXME: Does `hdr' really mean `host'? No.
2376 if (FD_ISSET(csp->sfd, &rfds))
2379 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2383 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2385 if (http->ssl && (fwd->forward_host == NULL))
2388 * Just hang up. We already confirmed the client's CONNECT
2389 * request with status code 200 and unencrypted content is
2390 * no longer welcome.
2392 log_error(LOG_LEVEL_ERROR,
2393 "CONNECT already confirmed. Unable to tell the client about the problem.");
2396 else if (byte_count)
2399 * Just hang up. We already transmitted the original headers
2400 * and parts of the original content and therefore missed the
2401 * chance to send an error message (without risking data corruption).
2403 * XXX: we could retry with a fancy range request here.
2405 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2406 "Unable to tell the client about the problem.");
2410 rsp = error_response(csp, "connect-failed", errno);
2414 send_crunch_response(csp, rsp);
2420 /* Add a trailing zero. This lets filter_popups
2421 * use string operations.
2425 #ifdef FEATURE_KILL_POPUPS
2426 /* Filter the popups on this read. */
2427 if (block_popups_now)
2429 filter_popups(buf, csp);
2431 #endif /* def FEATURE_KILL_POPUPS */
2433 /* Normally, this would indicate that we've read
2434 * as much as the server has sent us and we can
2435 * close the client connection. However, Microsoft
2436 * in its wisdom has released IIS/5 with a bug that
2437 * prevents it from sending the trailing \r\n in
2438 * a 302 redirect header (and possibly other headers).
2439 * To work around this if we've haven't parsed
2440 * a full header we'll append a trailing \r\n
2441 * and see if this now generates a valid one.
2443 * This hack shouldn't have any impacts. If we've
2444 * already transmitted the header or if this is a
2445 * SSL connection, then we won't bother with this
2446 * hack. So we only work on partially received
2447 * headers. If we append a \r\n and this still
2448 * doesn't generate a valid header, then we won't
2449 * transmit anything to the client.
2454 if (server_body || http->ssl)
2457 * If we have been buffering up the document,
2458 * now is the time to apply content modification
2459 * and send the result to the client.
2463 p = execute_content_filter(csp, content_filter);
2465 * If the content filter fails, use the original
2466 * buffer and length.
2467 * (see p != NULL ? p : csp->iob->cur below)
2471 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2474 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2476 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2479 hdr = list_to_text(csp->headers);
2482 /* FIXME Should handle error properly */
2483 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2487 * Shouldn't happen because this was the second sed run
2488 * and tags are only created for the first one.
2490 assert(!crunch_response_triggered(csp, crunchers_all));
2492 if (write_socket(csp->cfd, hdr, strlen(hdr))
2493 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2495 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2505 break; /* "game over, man" */
2509 * This is NOT the body, so
2510 * Let's pretend the server just sent us a blank line.
2512 snprintf(buf, sizeof(buf), "\r\n");
2513 len = (int)strlen(buf);
2516 * Now, let the normal header parsing algorithm below do its
2517 * job. If it fails, we'll exit instead of continuing.
2524 * If this is an SSL connection or we're in the body
2525 * of the server document, just write it to the client,
2526 * unless we need to buffer the body for later content-filtering
2529 if (server_body || http->ssl)
2534 * If there is no memory left for buffering the content, or the buffer limit
2535 * has been reached, switch to non-filtering mode, i.e. make & write the
2536 * header, flush the iob and buf, and get out of the way.
2538 if (add_to_iob(csp, buf, len))
2543 log_error(LOG_LEVEL_ERROR, "Flushing header and buffers. Stepping back from filtering.");
2545 hdr = list_to_text(csp->headers);
2549 * Memory is too tight to even generate the header.
2550 * Send our static "Out-of-memory" page.
2552 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2553 rsp = cgi_error_memory();
2554 send_crunch_response(csp, rsp);
2558 hdrlen = strlen(hdr);
2560 if (write_socket(csp->cfd, hdr, hdrlen)
2561 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
2562 || (write_socket(csp->cfd, buf, (size_t)len)))
2564 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2571 * Reset the byte_count to the amount of bytes
2572 * we just flushed. len will be added a few lines below,
2573 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2575 byte_count = (size_t)flushed;
2577 content_filter = NULL;
2583 if (write_socket(csp->cfd, buf, (size_t)len))
2585 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2589 byte_count += (size_t)len;
2594 /* we're still looking for the end of the
2595 * server's header ... (does that make header
2596 * parsing an "out of body experience" ?
2600 * buffer up the data we just read. If that fails,
2601 * there's little we can do but send our static
2602 * out-of-memory page.
2604 if (add_to_iob(csp, buf, len))
2606 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2607 rsp = cgi_error_memory();
2608 send_crunch_response(csp, rsp);
2613 /* Convert iob into something sed() can digest */
2614 if (JB_ERR_PARSE == get_server_headers(csp))
2618 /* Well, we tried our MS IIS/5
2619 * hack and it didn't work.
2620 * The header is incomplete
2621 * and there isn't anything
2622 * we can do about it.
2628 /* Since we have to wait for
2629 * more from the server before
2630 * we can parse the headers
2631 * we just continue here.
2637 /* Did we actually get anything? */
2638 if (NULL == csp->headers->first)
2640 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2641 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2642 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2643 free_http_request(http);
2647 /* we have now received the entire header.
2648 * filter it and send the result to the client
2650 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2652 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2654 hdr = list_to_text(csp->headers);
2657 /* FIXME Should handle error properly */
2658 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2661 if (crunch_response_triggered(csp, crunchers_light))
2664 * One of the tags created by a server-header
2665 * tagger triggered a crunch. We already
2666 * delivered the crunch response to the client
2667 * and are done here after cleaning up.
2672 /* Buffer and pcrs filter this if appropriate. */
2674 if (!http->ssl) /* We talk plaintext */
2677 #ifdef FEATURE_KILL_POPUPS
2678 /* Start blocking popups if appropriate. */
2679 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2680 (csp->action->flags & ACTION_NO_POPUPS) != 0) /* Policy allows */
2682 block_popups_now = 1;
2684 * Filter the part of the body that came in the same read
2685 * as the last headers:
2687 filter_popups(csp->iob->cur, csp);
2689 #endif /* def FEATURE_KILL_POPUPS */
2690 content_filter = get_filter_function(csp);
2693 * Only write if we're not buffering for content modification
2695 if (!content_filter)
2697 /* write the server's (modified) header to
2698 * the client (along with anything else that
2699 * may be in the buffer)
2702 if (write_socket(csp->cfd, hdr, strlen(hdr))
2703 || ((len = flush_socket(csp->cfd, csp)) < 0))
2705 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2707 /* the write failed, so don't bother
2708 * mentioning it to the client...
2709 * it probably can't hear us anyway.
2715 byte_count += (size_t)len;
2718 /* we're finished with the server's header */
2723 /* If this was a MS IIS/5 hack then it means
2724 * the server has already closed the
2725 * connection. Nothing more to read. Time
2736 return; /* huh? we should never get here */
2739 if (csp->content_length == 0)
2742 * If Privoxy didn't recalculate the
2743 * Content-Lenght, byte_count is still
2746 csp->content_length = byte_count;
2749 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2750 csp->ip_addr_str, http->ocmd, csp->content_length);
2754 /*********************************************************************
2758 * Description : This is little more than chat. We only "serve" to
2759 * to close any socket that chat may have opened.
2762 * 1 : csp = Current client state (buffers, headers, etc...)
2766 *********************************************************************/
2768 void serve(struct client_state *csp)
2769 #else /* ifndef AMIGA */
2770 static void serve(struct client_state *csp)
2771 #endif /* def AMIGA */
2774 close_socket(csp->cfd);
2776 if (csp->sfd != JB_INVALID_SOCKET)
2778 close_socket(csp->sfd);
2781 csp->flags &= ~CSP_FLAG_ACTIVE;
2787 /*********************************************************************
2789 * Function : server_thread
2791 * Description : We only exist to call `serve' in a threaded environment.
2794 * 1 : data = Current client state (buffers, headers, etc...)
2796 * Returns : Always 0.
2798 *********************************************************************/
2799 static int32 server_thread(void *data)
2801 serve((struct client_state *) data);
2809 /*********************************************************************
2813 * Description : Print usage info & exit.
2815 * Parameters : Pointer to argv[0] for identifying ourselves
2819 *********************************************************************/
2820 static void usage(const char *myname)
2822 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2826 #endif /* defined(unix) */
2829 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2830 #endif /* defined(unix) */
2831 "[--version] [configfile]\n"
2832 "Aborting\n", myname);
2837 #endif /* defined(unix) */
2840 /*********************************************************************
2842 * Function : initialize_mutexes
2844 * Description : Prepares mutexes if mutex support is available.
2848 * Returns : Void, exits in case of errors.
2850 *********************************************************************/
2851 static void initialize_mutexes(void)
2855 #ifdef FEATURE_PTHREAD
2857 * Prepare global mutex semaphores
2859 err = pthread_mutex_init(&log_mutex, 0);
2861 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2864 * XXX: The assumptions below are a bit naive
2865 * and can cause locks that aren't necessary.
2867 * For example older FreeBSD versions (< 6.x?)
2868 * have no gethostbyname_r, but gethostbyname is
2871 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2872 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2873 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2875 * XXX: should we use a single mutex for
2876 * localtime() and gmtime() as well?
2878 #ifndef HAVE_GMTIME_R
2879 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2880 #endif /* ndef HAVE_GMTIME_R */
2882 #ifndef HAVE_LOCALTIME_R
2883 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2884 #endif /* ndef HAVE_GMTIME_R */
2887 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2888 #endif /* ndef HAVE_RANDOM */
2889 #endif /* FEATURE_PTHREAD */
2892 * TODO: mutex support for mingw32 would be swell.
2897 printf("Fatal error. Mutex initialization failed: %s.\n",
2906 /*********************************************************************
2910 * Description : Load the config file and start the listen loop.
2911 * This function is a lot more *sane* with the `load_config'
2912 * and `listen_loop' functions; although it stills does
2913 * a *little* too much for my taste.
2916 * 1 : argc = Number of parameters (including $0).
2917 * 2 : argv = Array of (char *)'s to the parameters.
2919 * Returns : 1 if : can't open config file, unrecognized directive,
2920 * stats requested in multi-thread mode, can't open the
2921 * log file, can't open the jar file, listen port is invalid,
2922 * any load fails, and can't bind port.
2924 * Else main never returns, the process must be signaled
2925 * to terminate execution. Or, on Windows, use the
2926 * "File", "Exit" menu option.
2928 *********************************************************************/
2930 int real_main(int argc, const char *argv[])
2932 int main(int argc, const char *argv[])
2936 unsigned int random_seed;
2938 struct passwd *pw = NULL;
2939 struct group *grp = NULL;
2942 char *pre_chroot_nslookup_to_load_resolver = NULL;
2949 #if !defined(_WIN32)
2957 * Parse the command line arguments
2959 * XXX: simply printing usage information in case of
2960 * invalid arguments isn't particular user friendly.
2962 while (++argc_pos < argc)
2965 /* Check to see if the service must be installed or uninstalled */
2966 if (strncmp(argv[argc_pos], "--install", 9) == 0)
2968 const char *pName = argv[argc_pos] + 9;
2971 exit( (install_service(pName)) ? 0 : 1 );
2973 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
2975 const char *pName = argv[argc_pos] + 11;
2978 exit((uninstall_service(pName)) ? 0 : 1);
2980 else if (strcmp(argv[argc_pos], "--service" ) == 0)
2982 bRunAsService = TRUE;
2983 w32_set_service_cwd();
2984 atexit(w32_service_exit_notify);
2987 #endif /* defined(_WIN32) */
2990 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2992 if (strcmp(argv[argc_pos], "--help") == 0)
2997 else if(strcmp(argv[argc_pos], "--version") == 0)
2999 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3005 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3010 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3012 if (++argc_pos == argc) usage(argv[0]);
3013 pidfile = strdup(argv[argc_pos]);
3016 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3018 if (++argc_pos == argc) usage(argv[argc_pos]);
3020 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3023 if (NULL == (grp = getgrnam(p)))
3025 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3029 if (NULL == (pw = getpwnam(argv[argc_pos])))
3031 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3034 if (p != NULL) *--p = '\0';
3037 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3039 if (++argc_pos == argc) usage(argv[0]);
3040 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3043 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3047 #endif /* defined(unix) */
3049 else if (argc_pos + 1 != argc)
3052 * This is neither the last command line
3053 * option, nor was it recognized before,
3054 * therefore it must be invalid.
3060 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3062 configfile = argv[argc_pos];
3065 } /* -END- while (more arguments) */
3068 if ( *configfile != '/' )
3070 char cwd[BUFFER_SIZE];
3072 size_t abs_file_size;
3074 /* make config-filename absolute here */
3075 if (NULL == getcwd(cwd, sizeof(cwd)))
3077 perror("failed to get current working directory");
3082 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3083 basedir = strdup(cwd);
3085 if (NULL == basedir ||
3086 NULL == (abs_file = malloc(abs_file_size)))
3088 perror("malloc failed");
3091 strlcpy(abs_file, basedir, abs_file_size);
3092 strlcat(abs_file, "/", abs_file_size );
3093 strlcat(abs_file, configfile, abs_file_size);
3094 configfile = abs_file;
3096 #endif /* defined unix */
3100 clients->next = NULL;
3104 #elif defined(_WIN32)
3108 /* Prepare mutexes if supported and necessary. */
3109 initialize_mutexes();
3111 random_seed = (unsigned int)time(NULL);
3113 srandom(random_seed);
3116 #endif /* ifdef HAVE_RANDOM */
3119 * Unix signal handling
3121 * Catch the abort, interrupt and terminate signals for a graceful exit
3122 * Catch the hangup signal so the errlog can be reopened.
3123 * Ignore the broken pipe signals (FIXME: Why?)
3125 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3128 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3129 const int ignored_signals[] = { SIGPIPE, 0 };
3131 for (idx = 0; catched_signals[idx] != 0; idx++)
3133 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3134 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3136 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3137 #endif /* ifdef sun */
3139 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3143 for (idx = 0; ignored_signals[idx] != 0; idx++)
3145 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3147 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3152 #else /* ifdef _WIN32 */
3153 # ifdef _WIN_CONSOLE
3155 * We *are* in a windows console app.
3156 * Print a verbose messages about FAQ's and such
3158 printf("%s", win32_blurb);
3159 # endif /* def _WIN_CONSOLE */
3160 #endif /* def _WIN32 */
3163 /* Initialize the CGI subsystem */
3164 cgi_init_error_messages();
3167 * If runnig on unix and without the --nodaemon
3168 * option, become a daemon. I.e. fork, detach
3169 * from tty and get process group leadership
3182 if ( pid < 0 ) /* error */
3187 else if ( pid != 0 ) /* parent */
3192 * must check for errors
3193 * child died due to missing files aso
3196 wpid = waitpid( pid, &status, WNOHANG );
3205 /* Should be more portable, but not as well tested */
3210 #else /* ndef __FreeBSD__ */
3212 #endif /* ndef __FreeBSD__ */
3213 fd = open("/dev/tty", O_RDONLY);
3216 /* no error check here */
3217 ioctl( fd, TIOCNOTTY,0 );
3221 /* FIXME: should close stderr (fd 2) here too, but the test
3223 * and load config file is done in listen_loop() and puts
3224 * some messages on stderr there.
3231 } /* -END- if (!no_daemon) */
3234 * As soon as we have written the PID file, we can switch
3235 * to the user and group ID indicated by the --user option
3241 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3243 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3249 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3251 /* Read the time zone file from /etc before doing chroot. */
3253 if (NULL != pre_chroot_nslookup_to_load_resolver
3254 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3256 /* Initialize resolver library. */
3257 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3259 if (chroot(pw->pw_dir) < 0)
3261 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3265 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3268 if (setuid(pw->pw_uid))
3270 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3274 char putenv_dummy[64];
3276 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3277 if (putenv(putenv_dummy) != 0)
3279 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3282 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3283 if (putenv(putenv_dummy) != 0)
3285 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3291 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3294 #endif /* defined unix */
3297 /* This will be FALSE unless the command line specified --service
3301 /* Yup, so now we must attempt to establish a connection
3302 * with the service dispatcher. This will only work if this
3303 * process was launched by the service control manager to
3304 * actually run as a service. If this isn't the case, i've
3305 * known it take around 30 seconds or so for the call to return.
3308 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3309 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3311 /* Service has run, and at this point is now being stopped, so just return */
3316 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3318 /* An error occurred. Usually it's because --service was wrongly specified
3319 * and we were unable to connect to the Service Control Dispatcher because
3320 * it wasn't expecting us and is therefore not listening.
3322 * For now, just continue below to call the listen_loop function.
3325 #endif /* def _WIN32 */
3335 /*********************************************************************
3337 * Function : bind_port_helper
3339 * Description : Bind the listen port. Handles logging, and aborts
3343 * 1 : config = Privoxy configuration. Specifies port
3346 * Returns : Port that was opened.
3348 *********************************************************************/
3349 static jb_socket bind_port_helper(struct configuration_spec * config)
3354 if ( (config->haddr != NULL)
3355 && (config->haddr[0] == '1')
3356 && (config->haddr[1] == '2')
3357 && (config->haddr[2] == '7')
3358 && (config->haddr[3] == '.') )
3360 log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only",
3363 else if (config->haddr == NULL)
3365 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3370 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3371 config->hport, config->haddr);
3374 result = bind_port(config->haddr, config->hport, &bfd);
3381 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3382 "There may be another Privoxy or some other "
3383 "proxy running on port %d",
3384 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3385 config->hport, config->hport);
3388 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3389 "The hostname is not resolvable",
3390 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3393 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3394 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3397 /* shouldn't get here */
3398 return JB_INVALID_SOCKET;
3401 config->need_bind = 0;
3408 /* Without this simple workaround we get this compiler warning from _beginthread
3409 * warning C4028: formal parameter 1 different from declaration
3411 void w32_service_listen_loop(void *p)
3415 #endif /* def _WIN32 */
3418 /*********************************************************************
3420 * Function : listen_loop
3422 * Description : bind the listen port and enter a "FOREVER" listening loop.
3428 *********************************************************************/
3429 static void listen_loop(void)
3431 struct client_state *csp = NULL;
3433 struct configuration_spec * config;
3435 config = load_config();
3437 bfd = bind_port_helper(config);
3439 #ifdef FEATURE_GRACEFUL_TERMINATION
3440 while (!g_terminate)
3445 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3446 while (waitpid(-1, NULL, WNOHANG) > 0)
3448 /* zombie children */
3450 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3453 * Free data that was used by died threads
3459 * Re-open the errlog after HUP signal
3461 if (received_hup_signal)
3463 init_error_log(Argv[0], config->logfile, config->debug);
3464 received_hup_signal = 0;
3469 #ifdef FEATURE_COOKIE_JAR
3471 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3472 * already open. I think unload_configfile() is not being run, which should do
3473 * this work. Until that can get resolved, we'll use this workaround.
3477 if (csp->config->jar)
3479 fclose(csp->config->jar);
3480 csp->config->jar = NULL;
3482 #endif /* FEATURE_COOKIE_JAR */
3483 #endif /* __OS2__ */
3485 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3487 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3491 csp->flags |= CSP_FLAG_ACTIVE;
3492 csp->sfd = JB_INVALID_SOCKET;
3494 csp->config = config = load_config();
3496 if ( config->need_bind )
3499 * Since we were listening to the "old port", we will not see
3500 * a "listen" param change until the next IJB request. So, at
3501 * least 1 more request must be made for us to find the new
3502 * setting. I am simply closing the old socket and binding the
3505 * Which-ever is correct, we will serve 1 more page via the
3506 * old settings. This should probably be a "show-proxy-args"
3507 * request. This should not be a so common of an operation
3508 * that this will hurt people's feelings.
3513 bfd = bind_port_helper(config);
3516 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3518 if (!accept_connection(csp, bfd))
3520 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3533 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3536 #ifdef FEATURE_TOGGLE
3537 if (global_toggle_state)
3539 csp->flags |= CSP_FLAG_TOGGLED_ON;
3541 #endif /* def FEATURE_TOGGLE */
3543 if (run_loader(csp))
3545 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3546 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3550 if (block_acl(NULL,csp))
3552 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3553 close_socket(csp->cfd);
3557 #endif /* def FEATURE_ACL */
3559 /* add it to the list of clients */
3560 csp->next = clients->next;
3561 clients->next = csp;
3563 if (config->multi_threaded)
3567 /* this is a switch () statment in the C preprocessor - ugh */
3568 #undef SELECTED_ONE_OPTION
3570 /* Use Pthreads in preference to native code */
3571 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3572 #define SELECTED_ONE_OPTION
3574 pthread_t the_thread;
3575 pthread_attr_t attrs;
3577 pthread_attr_init(&attrs);
3578 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3579 errno = pthread_create(&the_thread, &attrs,
3580 (void * (*)(void *))serve, csp);
3581 child_id = errno ? -1 : 0;
3582 pthread_attr_destroy(&attrs);
3586 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3587 #define SELECTED_ONE_OPTION
3588 child_id = _beginthread(
3589 (void (*)(void *))serve,
3594 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3595 #define SELECTED_ONE_OPTION
3596 child_id = _beginthread(
3597 (void(* _Optlink)(void*))serve,
3603 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3604 #define SELECTED_ONE_OPTION
3606 thread_id tid = spawn_thread
3607 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3609 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3611 child_id = (int) tid;
3620 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3621 #define SELECTED_ONE_OPTION
3622 csp->cfd = ReleaseSocket(csp->cfd, -1);
3625 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3626 NP_Output, Output(),
3627 NP_CloseOutput, FALSE,
3628 NP_Name, (ULONG)"privoxy child",
3632 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3633 NP_Output, Output(),
3634 NP_CloseOutput, FALSE,
3635 NP_Name, (ULONG)"privoxy child",
3636 NP_StackSize, 200*1024,
3642 ((struct Task *)child_id)->tc_UserData = csp;
3643 Signal((struct Task *)child_id, SIGF_SINGLE);
3648 #if !defined(SELECTED_ONE_OPTION)
3651 /* This block is only needed when using fork().
3652 * When using threads, the server thread was
3653 * created and run by the call to _beginthread().
3655 if (child_id == 0) /* child */
3658 #ifdef FEATURE_TOGGLE
3659 int inherited_toggle_state = global_toggle_state;
3660 #endif /* def FEATURE_TOGGLE */
3665 * If we've been toggled or we've blocked the request, tell Mom
3668 #ifdef FEATURE_TOGGLE
3669 if (inherited_toggle_state != global_toggle_state)
3671 rc |= RC_FLAG_TOGGLED;
3673 #endif /* def FEATURE_TOGGLE */
3675 #ifdef FEATURE_STATISTICS
3676 if (csp->flags & CSP_FLAG_REJECTED)
3678 rc |= RC_FLAG_BLOCKED;
3680 #endif /* ndef FEATURE_STATISTICS */
3684 else if (child_id > 0) /* parent */
3686 /* in a fork()'d environment, the parent's
3687 * copy of the client socket and the CSP
3691 #if !defined(_WIN32) && !defined(__CYGWIN__)
3693 wait( &child_status );
3696 * Evaluate child's return code: If the child has
3697 * - been toggled, toggle ourselves
3698 * - blocked its request, bump up the stats counter
3701 #ifdef FEATURE_TOGGLE
3702 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3704 global_toggle_state = !global_toggle_state;
3706 #endif /* def FEATURE_TOGGLE */
3708 #ifdef FEATURE_STATISTICS
3710 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3714 #endif /* def FEATURE_STATISTICS */
3716 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3717 close_socket(csp->cfd);
3718 csp->flags &= ~CSP_FLAG_ACTIVE;
3722 #undef SELECTED_ONE_OPTION
3723 /* end of cpp switch () */
3725 if (child_id < 0) /* failed */
3727 char buf[BUFFER_SIZE];
3729 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3731 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3733 write_socket(csp->cfd, buf, strlen(buf));
3734 close_socket(csp->cfd);
3735 csp->flags &= ~CSP_FLAG_ACTIVE;
3746 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3748 /* Clean up. Aim: free all memory (no leaks) */
3749 #ifdef FEATURE_GRACEFUL_TERMINATION
3751 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3753 unload_current_config_file();
3754 unload_current_actions_file();
3755 unload_current_re_filterfile();
3756 #ifdef FEATURE_TRUST
3757 unload_current_trust_file();
3760 if (config->multi_threaded)
3767 } while ((clients->next != NULL) && (--i > 0));
3771 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3782 #ifdef FEATURE_COOKIE_JAR
3783 if (NULL != config->jar)
3785 fclose(config->jar);
3789 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3790 /* Cleanup - remove taskbar icon etc. */
3795 #endif /* FEATURE_GRACEFUL_TERMINATION */