1 const char jcc_rcs[] = "$Id: jcc.c,v 1.94 2006/07/18 14:48:46 david__schmidt 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 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 1.94 2006/07/18 14:48:46 david__schmidt
37 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
38 * with what was really the latest development (the v_3_0_branch branch)
40 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
41 * Thanks to Jindrich Makovicka for a race condition fix for the log
42 * file. The race condition remains for non-pthread implementations.
43 * Reference patch #1175720.
45 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
46 * Add the ability to check jpeg images for invalid
47 * lengths of comment blocks. Defensive strategy
48 * against the exploit:
49 * Microsoft Security Bulletin MS04-028
50 * Buffer Overrun in JPEG Processing (GDI+) Could
51 * Allow Code Execution (833987)
52 * Enabled with +inspect-jpegs in actions files.
54 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
55 * - Fixed usage info for non-unix platforms
56 * - Fixed small cmdline parsing bug
58 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
59 * Diagnostics: Now preserve the returncode of pthread_create
60 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
62 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
63 * No longer ignore SIGCHLD. Fixes bug #769381
65 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
66 * Close jarfile on graceful exit, remove stray line
68 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
69 * Cosmetics: Killed a warning, a typo and an allocation left at exit
71 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
72 * No longer rely on non-POSIX.1 extensions of getcwd().
75 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
76 * Replaced setenv() by posix-compliant putenv()
77 * Thanks to Neil McCalden (nmcc AT users.sf.net).
79 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
80 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
82 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
83 * Clean-up and extension of improvements for forked mode:
84 * - Child's return code now consists of flags RC_FLAG_*
85 * - Reporting toggle to parent now properly #ifdef'ed
86 * - Children now report blocking to parent. This enables
87 * statistics in forked mode
89 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
90 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
91 * Children now report having been toggled through _exit(17), parents
92 * watch for that code and toggle themselves if found.
94 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
95 * Wrapping all *_r functions (the non-_r versions of them) with
96 * mutex semaphores for OSX. Hopefully this will take care of all
97 * of those pesky crash reports.
99 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
100 * Fixed two mostly harmless mem leaks
102 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
103 * Fix: Head of global clients list now initialized to NULL
105 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
106 * Added basic support for OPTIONS and TRACE HTTP methods:
107 * - New interceptor direct_response() added in chat().
108 * - sed() moved to earlier in the process, so that the
109 * Host: header is evaluated before actions and forwarding
112 * Revision 1.92 2002/05/08 16:00:46 oes
113 * Chat's buffer handling:
114 * - Fixed bug with unchecked out-of-mem conditions
115 * while reading client request & server headers
116 * - No longer predict if the buffer limit will be exceeded
117 * in the next read -- check add_to_iob's new
118 * return code. If buffer couldn't be extended
119 * (policy or out-of-mem) while
120 * - reading from client: abort
121 * - reading server headers: send error page
122 * - buffering server body for filter: flush,
123 * and if that fails: send error page
125 * Revision 1.91 2002/04/08 20:35:58 swa
128 * Revision 1.90 2002/04/02 14:57:28 oes
129 * Made sending wafers independent of FEATURE_COOKIE_JAR
131 * Revision 1.89 2002/03/31 17:18:59 jongfoster
132 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
134 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
135 * More compiler warning message maintenance
137 * Revision 1.87 2002/03/26 22:29:54 swa
138 * we have a new homepage!
140 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
141 * Workaround for closing the jarfile before load_config() comes around again
143 * Revision 1.85 2002/03/24 15:23:33 jongfoster
146 * Revision 1.84 2002/03/24 13:25:43 swa
147 * name change related issues
149 * Revision 1.83 2002/03/16 23:54:06 jongfoster
150 * Adding graceful termination feature, to help look for memory leaks.
151 * If you enable this (which, by design, has to be done by hand
152 * editing config.h) and then go to http://i.j.b/die, then the program
153 * will exit cleanly after the *next* request. It should free all the
154 * memory that was used.
156 * Revision 1.82 2002/03/13 00:27:05 jongfoster
159 * Revision 1.81 2002/03/12 01:42:50 oes
160 * Introduced modular filters
162 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
163 * OS/2 port maintenance:
164 * - Fixed EMX build - it had decayed a little
165 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
166 * substituted a memset for now.
168 * Revision 1.79 2002/03/09 20:03:52 jongfoster
169 * - Making various functions return int rather than size_t.
170 * (Undoing a recent change). Since size_t is unsigned on
171 * Windows, functions like read_socket that return -1 on
172 * error cannot return a size_t.
174 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
175 * crashes, and also frequently caused JB to jump to 100%
176 * CPU and stay there. (Because it thought it had just
177 * read ((unsigned)-1) == 4Gb of data...)
179 * - The signature of write_socket has changed, it now simply
180 * returns success=0/failure=nonzero.
182 * - Trying to get rid of a few warnings --with-debug on
183 * Windows, I've introduced a new type "jb_socket". This is
184 * used for the socket file descriptors. On Windows, this
185 * is SOCKET (a typedef for unsigned). Everywhere else, it's
186 * an int. The error value can't be -1 any more, so it's
187 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
188 * Windows it maps to the #define INVALID_SOCKET.)
190 * - The signature of bind_port has changed.
192 * Revision 1.78 2002/03/08 21:35:04 oes
193 * Added optional group supplement to --user option. Will now use default group of user if no group given
195 * Revision 1.77 2002/03/07 03:52:06 oes
196 * - Fixed compiler warnings etc
197 * - Improved handling of failed DNS lookups
199 * Revision 1.76 2002/03/06 22:54:35 jongfoster
200 * Automated function-comment nitpicking.
202 * Revision 1.75 2002/03/06 10:02:19 oes
203 * Fixed stupid bug when --user was not given
205 * Revision 1.74 2002/03/06 00:49:31 jongfoster
206 * Fixing warning on Windows
207 * Making #ifdefs that refer to the same variable consistently
208 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
210 * Revision 1.73 2002/03/05 23:57:30 hal9
211 * Stray character 's' on line 1618 was breaking build.
213 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
214 * - Re-enable OS/2 building after new parms were added
215 * - Fix false out of memory report when resolving CGI templates when no IP
216 * address is available of failed attempt (a la no such domain)
218 * Revision 1.71 2002/03/05 18:13:56 oes
219 * Added --user option
221 * Revision 1.70 2002/03/05 04:52:42 oes
222 * Deleted non-errlog debugging code
224 * Revision 1.69 2002/03/04 23:50:00 jongfoster
225 * Splitting off bind_port() call into bind_port_helper(), with
228 * Revision 1.68 2002/03/04 20:17:32 oes
231 * Revision 1.67 2002/03/04 18:18:57 oes
232 * - Removed _DEBUG mode
233 * - Cleand up cmdline parsing
234 * - Introduced --no-daemon, --pidfile options
235 * - Cleaned up signal handling:
236 * - Terminate cleanly on INT, TERM and ABRT
237 * - Schedule logfile for re-opening on HUP
238 * - Ignore CHLD and PIPE
239 * - Leave the rest with their default handlers
240 * - Uniform handler registration
241 * - Added usage() function
242 * - Played styleguide police
244 * Revision 1.66 2002/03/03 15:06:55 oes
245 * Re-enabled automatic config reloading
247 * Revision 1.65 2002/03/03 14:49:11 oes
248 * Fixed CLF logging: Now uses client's original HTTP request
250 * Revision 1.64 2002/03/03 09:18:03 joergs
251 * Made jumbjuster work on AmigaOS again.
253 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
254 * Clean up a little CRLF unpleasantness that suddenly appeared
256 * Revision 1.62 2002/02/20 23:17:23 jongfoster
257 * Detecting some out-of memory conditions and exiting with a log message.
259 * Revision 1.61 2002/01/17 21:01:52 jongfoster
260 * Moving all our URL and URL pattern parsing code to urlmatch.c.
262 * Revision 1.60 2001/12/30 14:07:32 steudten
263 * - Add signal handling (unix)
264 * - Add SIGHUP handler (unix)
265 * - Add creation of pidfile (unix)
266 * - Add action 'top' in rc file (RH)
267 * - Add entry 'SIGNALS' to manpage
268 * - Add exit message to logfile (unix)
270 * Revision 1.59 2001/12/13 14:07:18 oes
271 * Fixed Bug: 503 error page now sent OK
273 * Revision 1.58 2001/11/30 23:37:24 jongfoster
274 * Renaming the Win32 config file to config.txt - this is almost the
275 * same as the corresponding UNIX name "config"
277 * Revision 1.57 2001/11/16 00:47:43 jongfoster
278 * Changing the tty-disconnection code to use setsid().
280 * Revision 1.56 2001/11/13 20:20:54 jongfoster
281 * Tabs->spaces, fixing a bug with missing {} around an if()
283 * Revision 1.55 2001/11/13 20:14:53 jongfoster
284 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
286 * Revision 1.54 2001/11/07 00:03:14 steudten
287 * Give reliable return value if an error
288 * occurs not just 0 with new daemon mode.
290 * Revision 1.53 2001/11/05 21:41:43 steudten
291 * Add changes to be a real daemon just for unix os.
292 * (change cwd to /, detach from controlling tty, set
293 * process group and session leader to the own process.
295 * Add some fatal-error log message for failed malloc().
296 * Add '-d' if compiled with 'configure --with-debug' to
297 * enable debug output.
299 * Revision 1.52 2001/10/26 20:11:20 jongfoster
300 * Fixing type mismatch
302 * Revision 1.51 2001/10/26 17:38:28 oes
305 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
306 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
307 * threads to call select() simultaneously. So, it's time to do a real, live,
308 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
309 * (native). Both versions will work, but using __OS2__ offers multi-threading.
311 * Revision 1.49 2001/10/23 21:41:35 jongfoster
312 * Added call to initialize the (statically-allocated of course)
313 * "out of memory" CGI response.
315 * Revision 1.48 2001/10/10 19:56:46 jongfoster
316 * Moving some code that wasn't cookie-related out of an #ifdef
319 * Revision 1.47 2001/10/10 16:44:36 oes
320 * Added CONNECT destination port limitation check
322 * Revision 1.46 2001/10/08 15:17:41 oes
323 * Re-enabled SSL forwarding
325 * Revision 1.45 2001/10/07 15:42:11 oes
326 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
328 * Moved downgrading of the HTTP version from parse_http_request to
329 * chat(), since we can't decide if it is necessary before we have
330 * determined the actions for the URL. The HTTP command is now
331 * *always* re-built so the repairs need no longer be special-cased.
333 * filter_popups now gets a csp pointer so it can raise the new
334 * CSP_FLAG_MODIFIED flag.
338 * Added configurable size limit for the IOB. If the IOB grows so
339 * large that the next read would exceed the limit, the header
340 * is generated, and the header & unfiltered buffer are flushed
341 * to the client. Chat then continues in non-buffering,
342 * non-filtering body mode.
344 * Revision 1.44 2001/10/02 18:13:57 oes
347 * Revision 1.43 2001/10/02 15:32:13 oes
348 * Moved generation of hdr
350 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
351 * Cleaning up 2 compiler warnings on OS/2.
353 * Revision 1.41 2001/09/16 17:05:14 jongfoster
354 * Removing unused #include showarg.h
356 * Revision 1.40 2001/09/16 15:41:45 jongfoster
357 * Fixing signed/unsigned comparison warning.
359 * Revision 1.39 2001/09/16 13:21:27 jongfoster
360 * Changes to use new list functions.
362 * Revision 1.38 2001/09/16 13:01:46 jongfoster
363 * Removing redundant function call that zeroed zalloc()'d memory.
365 * Revision 1.37 2001/09/10 11:12:24 oes
366 * Deleted unused variable
368 * Revision 1.36 2001/09/10 10:56:15 oes
369 * Silenced compiler warnings
371 * Revision 1.35 2001/07/31 14:44:22 oes
372 * Deleted unused size parameter from filter_popups()
374 * Revision 1.34 2001/07/30 22:08:36 jongfoster
375 * Tidying up #defines:
376 * - All feature #defines are now of the form FEATURE_xxx
377 * - Permanently turned off WIN_GUI_EDIT
378 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
380 * Revision 1.33 2001/07/29 19:32:00 jongfoster
381 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
383 * Revision 1.32 2001/07/29 18:47:05 jongfoster
384 * Adding missing #include "loadcfg.h"
386 * Revision 1.31 2001/07/29 12:17:48 oes
387 * Applied pthread fix by Paul Lieverse
389 * Revision 1.30 2001/07/25 22:57:13 jongfoster
390 * __BEOS__ no longer overrides FEATURE_PTHREAD.
391 * This is because FEATURE_PTHREAD will soon be widely used, so I
392 * want to keep it simple.
394 * Revision 1.29 2001/07/24 12:47:06 oes
395 * Applied BeOS support update by Eugenia
397 * Revision 1.28 2001/07/23 13:26:12 oes
398 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
400 * Revision 1.27 2001/07/19 19:09:47 haroon
401 * - Added code to take care of the situation where while processing the first
402 * server response (which includes the server header), after finding the end
403 * of the headers we were not looking past the end of the headers for
404 * content modification. I enabled it for filter_popups.
405 * Someone else should look to see if other similar operations should be
406 * done to the discarded portion of the buffer.
408 * Note 2001/07/20: No, the other content modification mechanisms will process
409 * the whole iob later anyway. --oes
411 * Revision 1.26 2001/07/18 12:31:36 oes
414 * Revision 1.25 2001/07/15 19:43:49 jongfoster
415 * Supports POSIX threads.
416 * Also removed some unused #includes.
418 * Revision 1.24 2001/07/13 14:00:40 oes
419 * - Generic content modification scheme:
420 * Each feature has its own applicability flag that is set
421 * from csp->action->flags.
422 * Replaced the "filtering" int flag , by a function pointer
423 * "content_filter" to the function that will do the content
424 * modification. If it is != NULL, the document will be buffered
425 * and processed through *content_filter, which must set
426 * csp->content_length and return a modified copy of the body
427 * or return NULL (on failiure).
428 * - Changed csp->is_text to the more generic bitmap csp->content_type
429 * which can currently take the valued CT_TEXT or CT_GIF
431 * - Removed all #ifdef PCRS
433 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
434 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
435 * the `block_acl' call. This prevents a compilation error when the user
436 * does not wish to use the "ACL" feature.
438 * Revision 1.22 2001/06/29 21:45:41 oes
439 * Indentation, CRLF->LF, Tab-> Space
441 * Revision 1.21 2001/06/29 13:29:36 oes
442 * - Cleaned up, improved comments
443 * - Unified all possible interceptors (CGI,
444 * block, trust, fast_redirect) in one
445 * place, with one (CGI) answer generation
446 * mechansim. Much clearer now.
447 * - Removed the GIF image generation, which
448 * is now done in filters.c:block_url()
449 * - Made error conditions like domain lookup
450 * failiure or (various) problems while talking
451 * to the server use cgi.c:error_response()
452 * instead of generating HTML/HTTP in chat() (yuck!)
453 * - Removed logentry from cancelled commit
455 * Revision 1.20 2001/06/09 10:55:28 jongfoster
456 * Changing BUFSIZ ==> BUFFER_SIZE
458 * Revision 1.19 2001/06/07 23:12:52 jongfoster
459 * Replacing function pointer in struct gateway with a directly
460 * called function forwarded_connect().
461 * Replacing struct gateway with struct forward_spec
463 * Revision 1.18 2001/06/03 19:12:16 oes
464 * introduced new cgi handling
466 * Revision 1.17 2001/06/01 20:07:23 jongfoster
467 * Now uses action +image-blocker{} rather than config->tinygif
469 * Revision 1.16 2001/06/01 18:49:17 jongfoster
470 * Replaced "list_share" with "list" - the tiny memory gain was not
471 * worth the extra complexity.
473 * Revision 1.15 2001/05/31 21:24:47 jongfoster
474 * Changed "permission" to "action" throughout.
475 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
476 * Moved vanilla wafer check into chat(), since we must now
477 * decide whether or not to add it based on the URL.
479 * Revision 1.14 2001/05/29 20:14:01 joergs
480 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
483 * Revision 1.13 2001/05/29 09:50:24 jongfoster
484 * Unified blocklist/imagelist/permissionslist.
485 * File format is still under discussion, but the internal changes
488 * Also modified interceptor behaviour:
489 * - We now intercept all URLs beginning with one of the following
490 * prefixes (and *only* these prefixes):
492 * * http://ijbswa.sf.net/config/
493 * * http://ijbswa.sourceforge.net/config/
494 * - New interceptors "home page" - go to http://i.j.b/ to see it.
495 * - Internal changes so that intercepted and fast redirect pages
496 * are not replaced with an image.
497 * - Interceptors now have the option to send a binary page direct
498 * to the client. (i.e. ijb-send-banner uses this)
499 * - Implemented show-url-info interceptor. (Which is why I needed
500 * the above interceptors changes - a typical URL is
501 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
502 * The previous mechanism would not have intercepted that, and
503 * if it had been intercepted then it then it would have replaced
506 * Revision 1.12 2001/05/27 22:17:04 oes
508 * - re_process_buffer no longer writes the modified buffer
509 * to the client, which was very ugly. It now returns the
510 * buffer, which it is then written by chat.
512 * - content_length now adjusts the Content-Length: header
513 * for modified documents rather than crunch()ing it.
514 * (Length info in csp->content_length, which is 0 for
515 * unmodified documents)
517 * - For this to work, sed() is called twice when filtering.
519 * Revision 1.11 2001/05/26 17:27:53 jongfoster
520 * Added support for CLF and fixed LOG_LEVEL_LOG.
521 * Also did CRLF->LF fix of my previous patch.
523 * Revision 1.10 2001/05/26 15:26:15 jongfoster
524 * ACL feature now provides more security by immediately dropping
525 * connections from untrusted hosts.
527 * Revision 1.9 2001/05/26 00:28:36 jongfoster
528 * Automatic reloading of config file.
529 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
530 * Most of the global variables have been moved to a new
531 * struct configuration_spec, accessed through csp->config->globalname
532 * Most of the globals remaining are used by the Win32 GUI.
534 * Revision 1.8 2001/05/25 22:43:18 jongfoster
535 * Fixing minor memory leak and buffer overflow.
537 * Revision 1.7 2001/05/25 22:34:30 jongfoster
540 * Revision 1.6 2001/05/23 00:13:58 joergs
541 * AmigaOS support fixed.
543 * Revision 1.5 2001/05/22 18:46:04 oes
545 * - Enabled filtering banners by size rather than URL
546 * by adding patterns that replace all standard banner
547 * sizes with the "Junkbuster" gif to the re_filterfile
549 * - Enabled filtering WebBugs by providing a pattern
550 * which kills all 1x1 images
552 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
553 * which is selected by the (nonstandard and therefore
554 * capital) letter 'U' in the option string.
555 * It causes the quantifiers to be ungreedy by default.
556 * Appending a ? turns back to greedy (!).
558 * - Added a new interceptor ijb-send-banner, which
559 * sends back the "Junkbuster" gif. Without imagelist or
560 * MSIE detection support, or if tinygif = 1, or the
561 * URL isn't recognized as an imageurl, a lame HTML
562 * explanation is sent instead.
564 * - Added new feature, which permits blocking remote
565 * script redirects and firing back a local redirect
567 * The feature is conditionally compiled, i.e. it
568 * can be disabled with --disable-fast-redirects,
569 * plus it must be activated by a "fast-redirects"
570 * line in the config file, has its own log level
571 * and of course wants to be displayed by show-proxy-args
572 * Note: Boy, all the #ifdefs in 1001 locations and
573 * all the fumbling with configure.in and acconfig.h
574 * were *way* more work than the feature itself :-(
576 * - Because a generic redirect template was needed for
577 * this, tinygif = 3 now uses the same.
579 * - Moved GIFs, and other static HTTP response templates
584 * - Removed some >400 CRs again (Jon, you really worked
587 * Revision 1.4 2001/05/21 19:34:01 jongfoster
588 * Made failure to bind() a fatal error.
590 * Revision 1.3 2001/05/20 01:21:20 jongfoster
591 * Version 2.9.4 checkin.
592 * - Merged popupfile and cookiefile, and added control over PCRS
593 * filtering, in new "permissionsfile".
594 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
595 * file error you now get a message box (in the Win32 GUI) rather
596 * than the program exiting with no explanation.
597 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
599 * - Removed tabs from "config"
600 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
601 * - Bumped up version number.
603 * Revision 1.2 2001/05/17 22:34:44 oes
604 * - Added hint on GIF char array generation to jcc.c
605 * - Cleaned CRLF's from the sources and related files
606 * - Repaired logging for REF and FRC
608 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
609 * Initial import of version 2.9.3 source tree
612 *********************************************************************/
618 #include <sys/types.h>
625 #ifdef FEATURE_PTHREAD
627 #endif /* def FEATURE_PTHREAD */
630 # ifndef FEATURE_PTHREAD
634 # include <windows.h>
635 # include <process.h>
636 # endif /* ndef FEATURE_PTHREAD */
639 # ifndef _WIN_CONSOLE
641 # endif /* ndef _WIN_CONSOLE */
643 #else /* ifndef _WIN32 */
645 # if !defined (__OS2__)
647 # include <sys/wait.h>
648 # endif /* ndef __OS2__ */
649 # include <sys/time.h>
650 # include <sys/stat.h>
651 # include <sys/ioctl.h>
654 #include <sys/termios.h>
665 # include <socket.h> /* BeOS has select() for sockets only. */
666 # include <OS.h> /* declarations for threads and stuff. */
669 # if defined(__EMX__) || defined(__OS2__)
670 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
675 #define bzero(B,N) memset(B,0x00,n)
690 #include "killpopup.h"
691 #include "miscutil.h"
693 #include "jbsockets.h"
698 #include "urlmatch.h"
700 const char jcc_h_rcs[] = JCC_H_VERSION;
701 const char project_h_rcs[] = PROJECT_H_VERSION;
704 struct client_state clients[1];
705 struct file_list files[1];
707 #ifdef FEATURE_STATISTICS
708 int urls_read = 0; /* total nr of urls read inc rejected */
709 int urls_rejected = 0; /* total nr of urls rejected */
710 #endif /* def FEATURE_STATISTICS */
712 #ifdef FEATURE_GRACEFUL_TERMINATION
716 static void listen_loop(void);
717 static void chat(struct client_state *csp);
719 void serve(struct client_state *csp);
720 #else /* ifndef AMIGA */
721 static void serve(struct client_state *csp);
722 #endif /* def AMIGA */
725 static int32 server_thread(void *data);
726 #endif /* def __BEOS__ */
729 #define sleep(N) Sleep(((N) * 1000))
733 #define sleep(N) DosSleep(((N) * 100))
738 * Hit OSX over the head with a hammer. Protect all *_r functions.
740 pthread_mutex_t gmtime_mutex;
741 pthread_mutex_t localtime_mutex;
742 pthread_mutex_t gethostbyaddr_mutex;
743 pthread_mutex_t gethostbyname_mutex;
744 #endif /* def OSX_DARWIN */
746 #ifdef FEATURE_PTHREAD
747 pthread_mutex_t log_mutex;
748 pthread_mutex_t log_init_mutex;
749 #endif /* FEATURE_PTHREAD */
751 #if defined(unix) || defined(__EMX__)
752 const char *basedir = NULL;
753 const char *pidfile = NULL;
754 int received_hup_signal = 0;
755 #endif /* defined unix */
757 /* The vanilla wafer. */
758 static const char VANILLA_WAFER[] =
759 "NOTICE=TO_WHOM_IT_MAY_CONCERN_"
760 "Do_not_send_me_any_copyrighted_information_other_than_the_"
761 "document_that_I_am_requesting_or_any_of_its_necessary_components._"
762 "In_particular_do_not_send_me_any_cookies_that_"
763 "are_subject_to_a_claim_of_copyright_by_anybody._"
764 "Take_notice_that_I_refuse_to_be_bound_by_any_license_condition_"
765 "(copyright_or_otherwise)_applying_to_any_cookie._";
768 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
769 /*********************************************************************
771 * Function : sig_handler
773 * Description : Signal handler for different signals.
774 * Exit gracefully on ABRT, TERM and INT
775 * or set a flag that will cause the errlog
776 * to be reopened by the main thread on HUP.
779 * 1 : the_signal = the signal cause this function to call
783 *********************************************************************/
784 static void sig_handler(int the_signal)
791 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
799 received_hup_signal = 1;
804 * We shouldn't be here, unless we catch signals
805 * in main() that we can't handle here!
807 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
815 /*********************************************************************
819 * Description : Once a connection to the client has been accepted,
820 * this function is called (via serve()) to handle the
821 * main business of the communication. When this
822 * function returns, the caller must close the client
826 * 1 : csp = Current client state (buffers, headers, etc...)
828 * Returns : On success, the number of bytes written are returned (zero
829 * indicates nothing was written). On error, -1 is returned,
830 * and errno is set appropriately. If count is zero and the
831 * file descriptor refers to a regular file, 0 will be
832 * returned without causing any other effect. For a special
833 * file, the results are not portable.
835 *********************************************************************/
836 static void chat(struct client_state *csp)
839 * This next lines are a little ugly, but they simplifies the if statements
840 * below. Basically if TOGGLE, then we want the if to test if the
841 * CSP_FLAG_TOGGLED_ON flag ist set, else we don't. And if FEATURE_FORCE_LOAD,
842 * then we want the if to test for CSP_FLAG_FORCED , else we don't
844 #ifdef FEATURE_TOGGLE
845 # define IS_TOGGLED_ON_AND (csp->flags & CSP_FLAG_TOGGLED_ON) &&
846 #else /* ifndef FEATURE_TOGGLE */
847 # define IS_TOGGLED_ON_AND
848 #endif /* ndef FEATURE_TOGGLE */
849 #ifdef FEATURE_FORCE_LOAD
850 # define IS_NOT_FORCED_AND !(csp->flags & CSP_FLAG_FORCED) &&
851 #else /* ifndef FEATURE_FORCE_LOAD */
852 # define IS_NOT_FORCED_AND
853 #endif /* def FEATURE_FORCE_LOAD */
855 #define IS_ENABLED_AND IS_TOGGLED_ON_AND IS_NOT_FORCED_AND
857 char buf[BUFFER_SIZE];
865 int ms_iis5_hack = 0;
867 unsigned int socks_retries = 0;
868 const struct forward_spec * fwd;
869 struct http_request *http;
870 int len; /* for buffer sizes */
871 #ifdef FEATURE_KILL_POPUPS
872 int block_popups; /* bool, 1==will block popups */
873 int block_popups_now = 0; /* bool, 1==currently blocking popups */
874 #endif /* def FEATURE_KILL_POPUPS */
876 int pcrs_filter; /* bool, 1==will filter through pcrs */
877 int gif_deanimate; /* bool, 1==will deanimate gifs */
878 int jpeg_inspect; /* bool, 1==will inspect jpegs */
880 /* Function that does the content filtering for the current request */
881 char *(*content_filter)() = NULL;
883 /* Skeleton for HTTP response, if we should intercept the request */
884 struct http_response *rsp;
889 * Read the client's request. Note that since we're not using select() we
890 * could get blocked here if a client connected, then didn't say anything!
895 len = read_socket(csp->cfd, buf, sizeof(buf));
897 if (len <= 0) break; /* error! */
900 * If there is no memory left for buffering the
901 * request, there is nothing we can do but hang up
903 if (add_to_iob(csp, buf, len))
908 req = get_header(csp);
912 break; /* no HTTP request! */
917 continue; /* more to come! */
920 #ifdef FEATURE_FORCE_LOAD
921 /* If this request contains the FORCE_PREFIX,
922 * better get rid of it now and set the force flag --oes
925 if (strstr(req, FORCE_PREFIX))
927 strclean(req, FORCE_PREFIX);
928 log_error(LOG_LEVEL_FORCE, "Enforcing request \"%s\".\n", req);
929 csp->flags |= CSP_FLAG_FORCED;
932 #endif /* def FEATURE_FORCE_LOAD */
934 parse_http_request(req, http, csp);
939 if (http->cmd == NULL)
941 strcpy(buf, CHEADER);
942 write_socket(csp->cfd, buf, strlen(buf));
944 log_error(LOG_LEVEL_CLF, "%s - - [%T] \" \" 400 0", csp->ip_addr_str);
946 free_http_request(http);
950 /* decide how to route the HTTP request */
952 if ((fwd = forward_url(http, csp)) == NULL)
954 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
955 /* Never get here - LOG_LEVEL_FATAL causes program exit */
958 /* build the http request to send to the server
959 * we have to do one of the following:
961 * create = use the original HTTP request to create a new
962 * HTTP request that has either the path component
963 * without the http://domainspec (w/path) or the
964 * full orininal URL (w/url)
965 * Note that the path and/or the HTTP version may
966 * have been altered by now.
968 * connect = Open a socket to the host:port of the server
969 * and short-circuit server and client socket.
971 * pass = Pass the request unchanged if forwarding a CONNECT
972 * request to a parent proxy. Note that we'll be sending
973 * the CFAIL message ourselves if connecting to the parent
974 * fails, but we won't send a CSUCCEED message if it works,
975 * since that would result in a double message (ours and the
976 * parent's). After sending the request to the parent, we simply
982 * +--------+--------+
984 * 0 | create | connect|
986 * Forwarding +--------+--------+
988 * 1 | create | pass |
990 * +--------+--------+
995 * Determine the actions for this URL
997 #ifdef FEATURE_TOGGLE
998 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
1000 /* Most compatible set of actions (i.e. none) */
1001 init_current_action(csp->action);
1004 #endif /* ndef FEATURE_TOGGLE */
1006 url_actions(http, csp);
1011 * Check if a CONNECT request is allowable:
1012 * In the absence of a +limit-connect action, allow only port 443.
1013 * If there is an action, allow whatever matches the specificaton.
1017 if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443)
1018 || (csp->action->flags & ACTION_LIMIT_CONNECT
1019 && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) )
1021 if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS)
1023 /* The response will violate the specs, but makes unblocking easier. */
1024 log_error(LOG_LEVEL_ERROR, "Marking suspicious CONNECT request from %s for blocking.",
1026 csp->action->flags |= ACTION_BLOCK;
1031 strcpy(buf, CFORBIDDEN);
1032 write_socket(csp->cfd, buf, strlen(buf));
1033 log_error(LOG_LEVEL_CONNECT, "Denying suspicious CONNECT request from %s", csp->ip_addr_str);
1034 log_error(LOG_LEVEL_CLF, "%s - - [%T] \" \" 403 0", csp->ip_addr_str);
1042 * Downgrade http version from 1.1 to 1.0 if +downgrade
1045 if ( (http->ssl == 0)
1046 && (!strcmpic(http->ver, "HTTP/1.1"))
1047 && (csp->action->flags & ACTION_DOWNGRADE))
1050 http->ver = strdup("HTTP/1.0");
1052 if (http->ver == NULL)
1054 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1059 * Save a copy of the original request for logging
1061 http->ocmd = strdup(http->cmd);
1063 if (http->ocmd == NULL)
1065 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
1069 * (Re)build the HTTP request for non-SSL requests.
1070 * If forwarding, use the whole URL, else, use only the path.
1076 http->cmd = strdup(http->gpc);
1077 string_append(&http->cmd, " ");
1079 if (fwd->forward_host)
1081 string_append(&http->cmd, http->url);
1085 string_append(&http->cmd, http->path);
1087 string_append(&http->cmd, " ");
1088 string_append(&http->cmd, http->ver);
1090 if (http->cmd == NULL)
1092 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1094 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", http->cmd);
1096 enlist(csp->headers, http->cmd);
1100 * If the user has not supplied any wafers, and the user has not
1101 * told us to suppress the vanilla wafer, then send the vanilla wafer.
1103 if (list_is_empty(csp->action->multi[ACTION_MULTI_WAFER])
1104 && ((csp->action->flags & ACTION_VANILLA_WAFER) != 0))
1106 enlist(csp->action->multi[ACTION_MULTI_WAFER], VANILLA_WAFER);
1110 #ifdef FEATURE_KILL_POPUPS
1111 block_popups = ((csp->action->flags & ACTION_NO_POPUPS) != 0);
1112 #endif /* def FEATURE_KILL_POPUPS */
1114 pcrs_filter = (csp->rlist != NULL) && /* There are expressions to be used */
1115 (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER]));
1117 gif_deanimate = ((csp->action->flags & ACTION_DEANIMATE) != 0);
1119 jpeg_inspect = ((csp->action->flags & ACTION_JPEG_INSPECT) != 0);
1121 /* grab the rest of the client's headers */
1125 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
1127 len = read_socket(csp->cfd, buf, sizeof(buf));
1130 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
1135 * If there is no memory left for buffering the
1136 * request, there is nothing we can do but hang up
1138 if (add_to_iob(csp, buf, len))
1145 if (p == NULL) break;
1147 enlist(csp->headers, p);
1152 * We have a request. Now, check to see if we need to
1153 * intercept it, i.e. If ..
1157 /* We may not forward the request by rfc2616 sect 14.31 */
1158 (NULL != (rsp = direct_response(csp)))
1160 /* or a CGI call was detected and answered */
1161 || (NULL != (rsp = dispatch_cgi(csp)))
1163 /* or we are enabled and... */
1164 || (IS_ENABLED_AND (
1166 /* ..the request was blocked */
1167 ( NULL != (rsp = block_url(csp)))
1169 /* ..or untrusted */
1170 #ifdef FEATURE_TRUST
1171 || ( NULL != (rsp = trust_url(csp)))
1172 #endif /* def FEATURE_TRUST */
1174 /* ..or a fast redirect kicked in */
1175 #ifdef FEATURE_FAST_REDIRECTS
1176 || (((csp->action->flags & ACTION_FAST_REDIRECTS) != 0) &&
1177 (NULL != (rsp = redirect_url(csp))))
1178 #endif /* def FEATURE_FAST_REDIRECTS */
1182 /* Write the answer to the client */
1183 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1184 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1186 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
1189 #ifdef FEATURE_STATISTICS
1190 /* Count as a rejected request */
1191 csp->flags |= CSP_FLAG_REJECTED;
1192 #endif /* def FEATURE_STATISTICS */
1194 /* Log (FIXME: All intercept reasons appear as "crunch" with Status 200) */
1195 log_error(LOG_LEVEL_GPC, "%s%s crunch!", http->hostport, http->path);
1196 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 3", csp->ip_addr_str, http->ocmd);
1198 /* Clean up and return */
1199 free_http_response(rsp);
1203 hdr = sed(client_patterns, add_client_headers, csp);
1206 /* FIXME Should handle error properly */
1207 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
1210 list_remove_all(csp->headers);
1212 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
1214 if (fwd->forward_host)
1216 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
1217 fwd->forward_host, fwd->forward_port, http->hostport);
1221 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
1224 /* here we connect to the server, gateway, or the forwarder */
1226 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
1227 && (errno == EINVAL) && (socks_retries++ < 3))
1229 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
1230 socks_retries, http->hostport);
1233 if (csp->sfd == JB_INVALID_SOCKET)
1235 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
1238 if (errno == EINVAL)
1240 rsp = error_response(csp, "no-such-domain", errno);
1242 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 404 0",
1243 csp->ip_addr_str, http->ocmd);
1247 rsp = error_response(csp, "connect-failed", errno);
1249 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 503 0",
1250 csp->ip_addr_str, http->ocmd);
1254 /* Write the answer to the client */
1257 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1258 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1260 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
1264 free_http_response(rsp);
1269 log_error(LOG_LEVEL_CONNECT, "OK");
1271 if (fwd->forward_host || (http->ssl == 0))
1273 /* write the client's (modified) header to the server
1274 * (along with anything else that may be in the buffer)
1277 if (write_socket(csp->sfd, hdr, strlen(hdr))
1278 || (flush_socket(csp->sfd, csp) < 0))
1280 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
1283 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 503 0",
1284 csp->ip_addr_str, http->ocmd);
1286 rsp = error_response(csp, "connect-failed", errno);
1290 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1291 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1293 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
1297 free_http_response(rsp);
1305 * We're running an SSL tunnel and we're not forwarding,
1306 * so just send the "connect succeeded" message to the
1307 * client, flush the rest, and get out of the way.
1309 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 2\n",
1310 csp->ip_addr_str, http->ocmd);
1312 if (write_socket(csp->cfd, CSUCCEED, sizeof(CSUCCEED)-1))
1320 /* we're finished with the client's header */
1323 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
1325 /* pass data between the client and server
1326 * until one or the other shuts down the connection.
1335 * FD_ZERO here seems to point to an errant macro which crashes.
1336 * So do this by hand for now...
1338 memset(&rfds,0x00,sizeof(fd_set));
1342 FD_SET(csp->cfd, &rfds);
1343 FD_SET(csp->sfd, &rfds);
1345 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
1349 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
1353 /* this is the body of the browser's request
1354 * just read it and write it.
1357 if (FD_ISSET(csp->cfd, &rfds))
1359 len = read_socket(csp->cfd, buf, sizeof(buf));
1363 break; /* "game over, man" */
1366 if (write_socket(csp->sfd, buf, (size_t)len))
1368 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
1375 * The server wants to talk. It could be the header or the body.
1376 * If `hdr' is null, then it's the header otherwise it's the body.
1377 * FIXME: Does `hdr' really mean `host'? No.
1381 if (FD_ISSET(csp->sfd, &rfds))
1384 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
1388 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
1390 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 503 0",
1391 csp->ip_addr_str, http->ocmd);
1393 rsp = error_response(csp, "connect-failed", errno);
1397 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1398 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1400 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
1404 free_http_response(rsp);
1408 /* Add a trailing zero. This lets filter_popups
1409 * use string operations.
1413 #ifdef FEATURE_KILL_POPUPS
1414 /* Filter the popups on this read. */
1415 if (block_popups_now)
1417 filter_popups(buf, csp);
1419 #endif /* def FEATURE_KILL_POPUPS */
1421 /* Normally, this would indicate that we've read
1422 * as much as the server has sent us and we can
1423 * close the client connection. However, Microsoft
1424 * in its wisdom has released IIS/5 with a bug that
1425 * prevents it from sending the trailing \r\n in
1426 * a 302 redirect header (and possibly other headers).
1427 * To work around this if we've haven't parsed
1428 * a full header we'll append a trailing \r\n
1429 * and see if this now generates a valid one.
1431 * This hack shouldn't have any impacts. If we've
1432 * already transmitted the header or if this is a
1433 * SSL connection, then we won't bother with this
1434 * hack. So we only work on partially received
1435 * headers. If we append a \r\n and this still
1436 * doesn't generate a valid header, then we won't
1437 * transmit anything to the client.
1442 if (server_body || http->ssl)
1445 * If we have been buffering up the document,
1446 * now is the time to apply content modification
1447 * and send the result to the client.
1452 * If the content filter fails, use the original
1453 * buffer and length.
1454 * (see p != NULL ? p : csp->iob->cur below)
1456 if (NULL == (p = (*content_filter)(csp)))
1458 csp->content_length = csp->iob->eod - csp->iob->cur;
1461 hdr = sed(server_patterns_light, NULL, csp);
1465 /* FIXME Should handle error properly */
1466 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
1469 if (write_socket(csp->cfd, hdr, strlen(hdr))
1470 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
1472 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
1482 break; /* "game over, man" */
1486 * This is NOT the body, so
1487 * Let's pretend the server just sent us a blank line.
1489 len = sprintf(buf, "\r\n");
1492 * Now, let the normal header parsing algorithm below do its
1493 * job. If it fails, we'll exit instead of continuing.
1500 * If this is an SSL connection or we're in the body
1501 * of the server document, just write it to the client,
1502 * unless we need to buffer the body for later content-filtering
1505 if (server_body || http->ssl)
1510 * If there is no memory left for buffering the content, or the buffer limit
1511 * has been reached, switch to non-filtering mode, i.e. make & write the
1512 * header, flush the iob and buf, and get out of the way.
1514 if (add_to_iob(csp, buf, len))
1519 log_error(LOG_LEVEL_ERROR, "Flushing header and buffers. Stepping back from filtering.");
1521 hdr = sed(server_patterns, add_server_headers, csp);
1525 * Memory is too tight to even generate the header.
1526 * Send our static "Out-of-memory" page.
1528 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
1529 rsp = cgi_error_memory();
1531 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1532 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1534 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
1539 hdrlen = strlen(hdr);
1541 if (write_socket(csp->cfd, hdr, hdrlen)
1542 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
1543 || (write_socket(csp->cfd, buf, (size_t) len)))
1545 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
1551 byte_count += hdrlen + flushed + len;
1553 content_filter = NULL;
1560 if (write_socket(csp->cfd, buf, (size_t)len))
1562 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
1571 /* we're still looking for the end of the
1572 * server's header ... (does that make header
1573 * parsing an "out of body experience" ?
1577 * buffer up the data we just read. If that fails,
1578 * there's little we can do but send our static
1579 * out-of-memory page.
1581 if (add_to_iob(csp, buf, len))
1583 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
1584 rsp = cgi_error_memory();
1586 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1587 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1589 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
1594 /* get header lines from the iob */
1596 while ((p = get_header(csp)) != NULL)
1600 /* see following note */
1603 enlist(csp->headers, p);
1607 /* NOTE: there are no "empty" headers so
1608 * if the pointer `p' is not NULL we must
1609 * assume that we reached the end of the
1610 * buffer before we hit the end of the header.
1617 /* Well, we tried our MS IIS/5
1618 * hack and it didn't work.
1619 * The header is incomplete
1620 * and there isn't anything
1621 * we can do about it.
1627 /* Since we have to wait for
1628 * more from the server before
1629 * we can parse the headers
1630 * we just continue here.
1636 /* we have now received the entire header.
1637 * filter it and send the result to the client
1640 hdr = sed(server_patterns, add_server_headers, csp);
1643 /* FIXME Should handle error properly */
1644 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
1647 #ifdef FEATURE_KILL_POPUPS
1648 /* Start blocking popups if appropriate. */
1650 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
1651 !http->ssl && /* We talk plaintext */
1652 block_popups) /* Policy allows */
1654 block_popups_now = 1;
1656 * Filter the part of the body that came in the same read
1657 * as the last headers:
1659 filter_popups(csp->iob->cur, csp);
1662 #endif /* def FEATURE_KILL_POPUPS */
1664 /* Buffer and pcrs filter this if appropriate. */
1666 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
1667 !http->ssl && /* We talk plaintext */
1668 pcrs_filter) /* Policy allows */
1670 content_filter = pcrs_filter_response;
1673 /* Buffer and gif_deanimate this if appropriate. */
1675 if ((csp->content_type & CT_GIF) && /* It's an image/gif MIME-Type */
1676 !http->ssl && /* We talk plaintext */
1677 gif_deanimate) /* Policy allows */
1679 content_filter = gif_deanimate_response;
1682 /* Buffer and jpg_inspect this if appropriate. */
1684 if ((csp->content_type & CT_JPEG) && /* It's an image/jpeg MIME-Type */
1685 !http->ssl && /* We talk plaintext */
1686 jpeg_inspect) /* Policy allows */
1688 content_filter = jpeg_inspect_response;
1692 * Only write if we're not buffering for content modification
1694 if (!content_filter)
1696 /* write the server's (modified) header to
1697 * the client (along with anything else that
1698 * may be in the buffer)
1701 if (write_socket(csp->cfd, hdr, strlen(hdr))
1702 || ((len = flush_socket(csp->cfd, csp)) < 0))
1704 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
1706 /* the write failed, so don't bother
1707 * mentioning it to the client...
1708 * it probably can't hear us anyway.
1717 /* we're finished with the server's header */
1722 /* If this was a MS IIS/5 hack then it means
1723 * the server has already closed the
1724 * connection. Nothing more to read. Time
1735 return; /* huh? we should never get here */
1738 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
1739 csp->ip_addr_str, http->ocmd, byte_count);
1743 /*********************************************************************
1747 * Description : This is little more than chat. We only "serve" to
1748 * to close any socket that chat may have opened.
1751 * 1 : csp = Current client state (buffers, headers, etc...)
1755 *********************************************************************/
1757 void serve(struct client_state *csp)
1758 #else /* ifndef AMIGA */
1759 static void serve(struct client_state *csp)
1760 #endif /* def AMIGA */
1763 close_socket(csp->cfd);
1765 if (csp->sfd != JB_INVALID_SOCKET)
1767 close_socket(csp->sfd);
1770 csp->flags &= ~CSP_FLAG_ACTIVE;
1776 /*********************************************************************
1778 * Function : server_thread
1780 * Description : We only exist to call `serve' in a threaded environment.
1783 * 1 : data = Current client state (buffers, headers, etc...)
1785 * Returns : Always 0.
1787 *********************************************************************/
1788 static int32 server_thread(void *data)
1790 serve((struct client_state *) data);
1797 /*********************************************************************
1801 * Description : Print usage info & exit.
1803 * Parameters : Pointer to argv[0] for identifying ourselves
1807 *********************************************************************/
1808 void usage(const char *myname)
1810 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
1812 "Usage: %s [--help] [--version] [configfile]\n"
1814 "Usage: %s [--help] [--version] [--no-daemon] [--pidfile pidfile] [--user user[.group]] [configfile]\n"
1816 "Aborting.\n", myname);
1823 /*********************************************************************
1827 * Description : Load the config file and start the listen loop.
1828 * This function is a lot more *sane* with the `load_config'
1829 * and `listen_loop' functions; although it stills does
1830 * a *little* too much for my taste.
1833 * 1 : argc = Number of parameters (including $0).
1834 * 2 : argv = Array of (char *)'s to the parameters.
1836 * Returns : 1 if : can't open config file, unrecognized directive,
1837 * stats requested in multi-thread mode, can't open the
1838 * log file, can't open the jar file, listen port is invalid,
1839 * any load fails, and can't bind port.
1841 * Else main never returns, the process must be signaled
1842 * to terminate execution. Or, on Windows, use the
1843 * "File", "Exit" menu option.
1845 *********************************************************************/
1847 int real_main(int argc, const char *argv[])
1849 int main(int argc, const char *argv[])
1854 struct passwd *pw = NULL;
1855 struct group *grp = NULL;
1864 #if !defined(_WIN32)
1872 * Parse the command line arguments
1874 while (++argc_pos < argc)
1876 if (strcmp(argv[argc_pos], "--help") == 0)
1881 else if(strcmp(argv[argc_pos], "--version") == 0)
1883 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
1889 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
1894 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
1896 if (++argc_pos == argc) usage(argv[0]);
1897 pidfile = strdup(argv[argc_pos]);
1900 else if (strcmp(argv[argc_pos], "--user" ) == 0)
1902 if (++argc_pos == argc) usage(argv[argc_pos]);
1904 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
1907 if (NULL == (grp = getgrnam(p)))
1909 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
1913 if (NULL == (pw = getpwnam(argv[argc_pos])))
1915 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
1918 if (p != NULL) *--p = '\0';
1921 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
1926 #endif /* defined(unix) */
1930 configfile = argv[argc_pos];
1933 } /* -END- while (more arguments) */
1936 if ( *configfile != '/' )
1938 char *abs_file, cwd[1024];
1940 /* make config-filename absolute here */
1941 if ( !(getcwd(cwd, sizeof(cwd))))
1943 perror("get working dir failed");
1947 if (!(basedir = strdup(cwd))
1948 || (!(abs_file = malloc( strlen( basedir ) + strlen( configfile ) + 5 ))))
1950 perror("malloc failed");
1953 strcpy( abs_file, basedir );
1954 strcat( abs_file, "/" );
1955 strcat( abs_file, configfile );
1956 configfile = abs_file;
1958 #endif /* defined unix */
1962 clients->next = NULL;
1966 #elif defined(_WIN32)
1972 * Prepare global mutex semaphores
1974 pthread_mutex_init(&gmtime_mutex,0);
1975 pthread_mutex_init(&localtime_mutex,0);
1976 pthread_mutex_init(&gethostbyaddr_mutex,0);
1977 pthread_mutex_init(&gethostbyname_mutex,0);
1978 #endif /* def OSX_DARWIN */
1980 #ifdef FEATURE_PTHREAD
1981 pthread_mutex_init(&log_mutex,0);
1982 pthread_mutex_init(&log_init_mutex,0);
1983 #endif /* FEATURE_PTHREAD */
1986 * Unix signal handling
1988 * Catch the abort, interrupt and terminate signals for a graceful exit
1989 * Catch the hangup signal so the errlog can be reopened.
1990 * Ignore the broken pipe signals (FIXME: Why?)
1992 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1995 const int catched_signals[] = { SIGABRT, SIGTERM, SIGINT, SIGHUP, 0 };
1996 const int ignored_signals[] = { SIGPIPE, 0 };
1998 for (idx = 0; catched_signals[idx] != 0; idx++)
2000 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
2002 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
2006 for (idx = 0; ignored_signals[idx] != 0; idx++)
2008 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
2010 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
2015 #else /* ifdef _WIN32 */
2016 # ifdef _WIN_CONSOLE
2018 * We *are* in a windows console app.
2019 * Print a verbose messages about FAQ's and such
2021 printf(win32_blurb);
2022 # endif /* def _WIN_CONSOLE */
2023 #endif /* def _WIN32 */
2026 /* Initialize the CGI subsystem */
2027 cgi_init_error_messages();
2030 * If runnig on unix and without the --nodaemon
2031 * option, become a daemon. I.e. fork, detach
2032 * from tty and get process group leadership
2045 if ( pid < 0 ) /* error */
2050 else if ( pid != 0 ) /* parent */
2055 * must check for errors
2056 * child died due to missing files aso
2059 wpid = waitpid( pid, &status, WNOHANG );
2068 /* Should be more portable, but not as well tested */
2073 #else /* ndef __FreeBSD__ */
2075 #endif /* ndef __FreeBSD__ */
2076 fd = open("/dev/tty", O_RDONLY);
2079 /* no error check here */
2080 ioctl( fd, TIOCNOTTY,0 );
2084 /* FIXME: should close stderr (fd 2) here too, but the test
2086 * and load config file is done in listen_loop() and puts
2087 * some messages on stderr there.
2094 } /* -END- if (!no_daemon) */
2097 * As soon as we have written the PID file, we can switch
2098 * to the user and group ID indicated by the --user option
2104 if (((NULL != grp) && setgid(grp->gr_gid)) || (setgid(pw->pw_gid)))
2106 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
2112 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
2114 if (chroot(pw->pw_dir) < 0)
2116 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
2120 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
2123 if (setuid(pw->pw_uid))
2125 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
2129 char putenv_dummy[64];
2131 strcpy(putenv_dummy, "HOME=/");
2132 if (putenv(putenv_dummy) != 0)
2134 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
2137 snprintf(putenv_dummy, 64, "USER=%s", pw->pw_name);
2138 if (putenv(putenv_dummy) != 0)
2140 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
2146 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
2149 #endif /* defined unix */
2159 /*********************************************************************
2161 * Function : bind_port_helper
2163 * Description : Bind the listen port. Handles logging, and aborts
2167 * 1 : config = Privoxy configuration. Specifies port
2170 * Returns : Port that was opened.
2172 *********************************************************************/
2173 static jb_socket bind_port_helper(struct configuration_spec * config)
2178 if ( (config->haddr != NULL)
2179 && (config->haddr[0] == '1')
2180 && (config->haddr[1] == '2')
2181 && (config->haddr[2] == '7')
2182 && (config->haddr[3] == '.') )
2184 log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only",
2187 else if (config->haddr == NULL)
2189 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
2194 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
2195 config->hport, config->haddr);
2198 result = bind_port(config->haddr, config->hport, &bfd);
2205 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
2206 "There may be another Privoxy or some other "
2207 "proxy running on port %d",
2208 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
2209 config->hport, config->hport);
2212 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
2213 "The hostname is not resolvable",
2214 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
2217 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
2218 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
2221 /* shouldn't get here */
2222 return JB_INVALID_SOCKET;
2225 config->need_bind = 0;
2231 /*********************************************************************
2233 * Function : listen_loop
2235 * Description : bind the listen port and enter a "FOREVER" listening loop.
2241 *********************************************************************/
2242 static void listen_loop(void)
2244 struct client_state *csp = NULL;
2246 struct configuration_spec * config;
2248 config = load_config();
2250 bfd = bind_port_helper(config);
2252 #ifdef FEATURE_GRACEFUL_TERMINATION
2253 while (!g_terminate)
2258 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
2259 while (waitpid(-1, NULL, WNOHANG) > 0)
2261 /* zombie children */
2263 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
2266 * Free data that was used by died threads
2272 * Re-open the errlog after HUP signal
2274 if (received_hup_signal)
2276 init_error_log(Argv[0], config->logfile, config->debug);
2277 received_hup_signal = 0;
2282 #ifdef FEATURE_COOKIE_JAR
2284 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
2285 * already open. I think unload_configfile() is not being run, which should do
2286 * this work. Until that can get resolved, we'll use this workaround.
2290 if (csp->config->jar)
2292 fclose(csp->config->jar);
2293 csp->config->jar = NULL;
2295 #endif /* FEATURE_COOKIE_JAR */
2296 #endif /* __OS2__ */
2298 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
2300 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
2304 csp->flags |= CSP_FLAG_ACTIVE;
2305 csp->sfd = JB_INVALID_SOCKET;
2307 csp->config = config = load_config();
2309 if ( config->need_bind )
2312 * Since we were listening to the "old port", we will not see
2313 * a "listen" param change until the next IJB request. So, at
2314 * least 1 more request must be made for us to find the new
2315 * setting. I am simply closing the old socket and binding the
2318 * Which-ever is correct, we will serve 1 more page via the
2319 * old settings. This should probably be a "show-proxy-args"
2320 * request. This should not be a so common of an operation
2321 * that this will hurt people's feelings.
2326 bfd = bind_port_helper(config);
2329 log_error(LOG_LEVEL_CONNECT, "accept connection ... ");
2331 if (!accept_connection(csp, bfd))
2333 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
2346 log_error(LOG_LEVEL_CONNECT, "OK");
2349 #ifdef FEATURE_TOGGLE
2350 if (global_toggle_state)
2352 csp->flags |= CSP_FLAG_TOGGLED_ON;
2354 #endif /* def FEATURE_TOGGLE */
2356 if (run_loader(csp))
2358 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
2359 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2363 if (block_acl(NULL,csp))
2365 log_error(LOG_LEVEL_CONNECT, "Connection dropped due to ACL");
2366 close_socket(csp->cfd);
2370 #endif /* def FEATURE_ACL */
2372 /* add it to the list of clients */
2373 csp->next = clients->next;
2374 clients->next = csp;
2376 if (config->multi_threaded)
2380 /* this is a switch () statment in the C preprocessor - ugh */
2381 #undef SELECTED_ONE_OPTION
2383 /* Use Pthreads in preference to native code */
2384 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
2385 #define SELECTED_ONE_OPTION
2387 pthread_t the_thread;
2388 pthread_attr_t attrs;
2390 pthread_attr_init(&attrs);
2391 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
2392 errno = pthread_create(&the_thread, &attrs,
2394 child_id = errno ? -1 : 0;
2395 pthread_attr_destroy(&attrs);
2399 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
2400 #define SELECTED_ONE_OPTION
2401 child_id = _beginthread(
2402 (void (*)(void *))serve,
2407 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
2408 #define SELECTED_ONE_OPTION
2409 child_id = _beginthread(
2410 (void(* _Optlink)(void*))serve,
2416 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
2417 #define SELECTED_ONE_OPTION
2419 thread_id tid = spawn_thread
2420 (server_thread, "server", B_NORMAL_PRIORITY, csp);
2422 if ((tid >= 0) && (resume_thread(tid) == B_OK))
2424 child_id = (int) tid;
2433 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
2434 #define SELECTED_ONE_OPTION
2435 csp->cfd = ReleaseSocket(csp->cfd, -1);
2437 if((child_id = (int)CreateNewProcTags(
2438 NP_Entry, (ULONG)server_thread,
2439 NP_Output, Output(),
2440 NP_CloseOutput, FALSE,
2441 NP_Name, (ULONG)"privoxy child",
2442 NP_StackSize, 200*1024,
2446 ((struct Task *)child_id)->tc_UserData = csp;
2447 Signal((struct Task *)child_id, SIGF_SINGLE);
2452 #if !defined(SELECTED_ONE_OPTION)
2455 /* This block is only needed when using fork().
2456 * When using threads, the server thread was
2457 * created and run by the call to _beginthread().
2459 if (child_id == 0) /* child */
2462 #ifdef FEATURE_TOGGLE
2463 int inherited_toggle_state = global_toggle_state;
2464 #endif /* def FEATURE_TOGGLE */
2469 * If we've been toggled or we've blocked the request, tell Mom
2472 #ifdef FEATURE_TOGGLE
2473 if (inherited_toggle_state != global_toggle_state)
2475 rc |= RC_FLAG_TOGGLED;
2477 #endif /* def FEATURE_TOGGLE */
2479 #ifdef FEATURE_STATISTICS
2480 if (csp->flags & CSP_FLAG_REJECTED)
2482 rc |= RC_FLAG_BLOCKED;
2484 #endif /* ndef FEATURE_STATISTICS */
2488 else if (child_id > 0) /* parent */
2490 /* in a fork()'d environment, the parent's
2491 * copy of the client socket and the CSP
2495 #if !defined(_WIN32) && !defined(__CYGWIN__)
2497 wait( &child_status );
2500 * Evaluate child's return code: If the child has
2501 * - been toggled, toggle ourselves
2502 * - blocked its request, bump up the stats counter
2505 #ifdef FEATURE_TOGGLE
2506 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
2508 global_toggle_state = !global_toggle_state;
2510 #endif /* def FEATURE_TOGGLE */
2512 #ifdef FEATURE_STATISTICS
2514 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
2518 #endif /* def FEATURE_STATISTICS */
2520 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
2521 close_socket(csp->cfd);
2522 csp->flags &= ~CSP_FLAG_ACTIVE;
2526 #undef SELECTED_ONE_OPTION
2527 /* end of cpp switch () */
2529 if (child_id < 0) /* failed */
2531 char buf[BUFFER_SIZE];
2533 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
2535 sprintf(buf , "Privoxy: can't fork: errno = %d", errno);
2537 write_socket(csp->cfd, buf, strlen(buf));
2538 close_socket(csp->cfd);
2539 csp->flags &= ~CSP_FLAG_ACTIVE;
2550 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
2552 /* Clean up. Aim: free all memory (no leaks) */
2553 #ifdef FEATURE_GRACEFUL_TERMINATION
2555 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
2557 unload_current_config_file();
2558 unload_current_actions_file();
2559 unload_current_re_filterfile();
2560 #ifdef FEATURE_TRUST
2561 unload_current_trust_file();
2564 if (config->multi_threaded)
2571 } while ((clients->next != NULL) && (--i > 0));
2575 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
2586 #ifdef FEATURE_COOKIE_JAR
2587 if (NULL != config->jar)
2589 fclose(config->jar);
2593 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
2594 /* Cleanup - remove taskbar icon etc. */
2599 #endif /* FEATURE_GRACEFUL_TERMINATION */