1 const char jcc_rcs[] = "$Id: jcc.c,v 1.204 2008/11/06 19:42:17 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) 2001-2008 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 1.204 2008/11/06 19:42:17 fabiankeil
37 * Fix last-chunk detection hack to also apply
38 * if buf[] contains nothing but the last-chunk.
40 * Revision 1.203 2008/11/06 18:34:35 fabiankeil
41 * Factor receive_client_request() and
42 * parse_client_request() out of chat().
44 * Revision 1.202 2008/11/02 18:40:34 fabiankeil
45 * If we received a different amount of data than we expected,
46 * log a warning and make sure the server socket isn't reused.
48 * Revision 1.201 2008/11/02 16:48:20 fabiankeil
49 * Revert revision 1.195 and try again.
51 * Revision 1.200 2008/10/26 16:53:18 fabiankeil
54 * Revision 1.199 2008/10/26 15:36:10 fabiankeil
55 * Remove two debug messages with LOG_LEVEL_INFO.
57 * Revision 1.198 2008/10/22 15:19:55 fabiankeil
58 * Once More, With Feeling: if there is no logfile
59 * because the user didn't specify one, we shouldn't
60 * call init_error_log() after receiving SIGHUP either.
62 * Revision 1.197 2008/10/20 17:02:40 fabiankeil
63 * If SIGHUP is received while we aren't running in daemon
64 * mode, calling init_error_log() would be a mistake.
66 * Revision 1.196 2008/10/16 09:16:41 fabiankeil
67 * - Fix two gcc44 conversion warnings.
68 * - Don't bother logging the last five bytes
71 * Revision 1.195 2008/10/13 16:04:37 fabiankeil
72 * Make sure we don't try to reuse tainted server sockets.
74 * Revision 1.194 2008/10/12 18:35:18 fabiankeil
75 * The last commit was a bit too ambitious, apparently the content
76 * length adjustment is only necessary if we aren't buffering.
78 * Revision 1.193 2008/10/12 15:57:35 fabiankeil
79 * Fix content length calculation if we read headers
80 * and the start of the body at once. Now that we have
81 * FEATURE_CONNECTION_KEEP_ALIVE, it actually matters.
83 * Revision 1.192 2008/10/11 18:19:14 fabiankeil
84 * Even more chat() cosmetics.
86 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
87 * Reformat some comments in chat().
89 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
90 * In case of chunk-encoded content, stop reading if
91 * the buffer looks like it ends with the last chunk.
93 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
94 * Let server_response_is_complete() deal properly with
95 * content that is neither buffered nor read all at once.
97 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
98 * Flush work-in-progress changes to keep outgoing connections
99 * alive where possible. Incomplete and mostly #ifdef'd out.
101 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
102 * Add mutex lock support for _WIN32.
104 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
105 * Prepare for critical sections on Windows by adding a
106 * layer of indirection before the pthread mutex functions.
108 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
109 * Remove FEATURE_COOKIE_JAR.
111 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
112 * - Silence LLVM/Clang complaint.
113 * - Make received_hup_signal static.
114 * - Hide definitions for basedir, pidfile and received_hup_signal
115 * from __EMX__ as they only seem to be used in case of #ifdef unix.
117 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
118 * Accept Shoutcast responses again. Problem reported
119 * and fix suggested by Stefan in #2062860.
121 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
122 * Fix possible NULL-pointer dereference reported
123 * by din_a4 in #2003937. Pointy hat to me.
125 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
126 * Streamline sed()'s prototype and declare
127 * the header parse and add structures static.
129 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
130 * - Mark csp as immutable for send_crunch_response().
131 * - Fix comment spelling.
133 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
134 * Factor update_server_headers() out of sed(), ditch the
135 * first_run hack and make server_patterns_light static.
137 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
138 * Don't provide get_header() with the whole client state
139 * structure when it only needs access to csp->iob.
141 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
142 * Make the "read the rest of the headers" loop a bit more readable.
144 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
145 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
146 * - Remove useless comment.
148 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
149 * Fix comment grammar.
151 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
152 * Remove the pointless buffer in client_protocol_is_unsupported().
154 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
155 * Least-effort fix for bug #1821930 (reported by Lee):
156 * If the response doesn't look like HTTP,
157 * tell the client and log the problem.
159 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
160 * Don't pass the whole csp structure to flush_socket()
161 * when it only needs a file descriptor and a buffer.
163 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
164 * Remove kill-popups action.
166 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
167 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
169 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
170 * Remove the treat-forbidden-connects-like-blocks action. We now
171 * use the "blocked" page for forbidden CONNECT requests by default.
173 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
174 * Also use shiny new connect_port_is_forbidden() in jcc.c.
176 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
177 * Rename url_actions() to get_url_actions() and let it
178 * use the standard parameter ordering.
180 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
181 * Let forward_url() use the standard parameter ordering
182 * and mark its second parameter immutable.
184 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
185 * Remove the "Listening ... for local connections only" log message.
186 * Whether or not remote connections are able to reach Privoxy is up
187 * to the operating system.
189 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
190 * Prevent the log messages for CONNECT requests to unacceptable
191 * ports from printing the limit-connect argument as [null] if
192 * limit-connect hasn't been explicitly enabled.
194 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
195 * Make sure all console-mode apps get a usage() instance
197 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
198 * Reword NO_SERVER_DATA_RESPONSE to make it harder
199 * to misunderstand what the message is all about.
201 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
202 * Unbreak trustfile which previously didn't work without
203 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
205 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
206 * Plug memory leak. Spotted by Valgrind, triggered by
207 * Privoxy-Regression-Test feeding proxyfuzz.py.
209 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
210 * In the HTTP snipplets, refer to the client as client.
212 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
213 * Emit a log message when activating the MS IIS5 hack.
215 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
216 * Log the "weak randomization factor" warning only
217 * once for mingw32 and provide some more details.
219 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
220 * Initialize log module after initializing mutexes, future
221 * deadlocks in that code should now work cross-platform.
223 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
224 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
225 * Reported by Bert van Leeuwen in BR#1818808.
227 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
228 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
230 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
231 * When in daemon mode, close stderr after the configuration file has been
232 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
234 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
235 * - Fix a crash when parsing invalid requests whose first header
236 * is rejected by get_header(). Regression (re?)introduced
237 * in r1.143 by yours truly.
238 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
239 * client_cookie_adder() to make sure send-vanilla-wafer can be
240 * controlled through tags (and thus regression-tested).
242 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
243 * - Move get_filter_function() from jcc.c to filters.c
244 * so the filter functions can be static.
245 * - Don't bother filtering body-less responses.
247 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
248 * Execute content filters through execute_content_filter().
250 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
251 * Initialize req to NULL to make sure it's defined if the
252 * first read_socket() call fails. Reported by icmp30.
254 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
255 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
256 * extensive comments moved to user manual.
258 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
259 * Don't crash if a broken header filter wiped out the request line.
261 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
262 * Fix byte_count calculation in case of flushes
263 * and don't parse the server headers a second time.
265 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
266 * - If there's a connection problem after we already forwarded
267 * parts of the original content, just hang up. Fixes BR#1776724.
268 * - Fix warnings about unused code on mingw32.
269 * - In case of flushes, calculate the byte count
270 * less incorrectly (I think).
272 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
273 * Add some more prototypes for static functions.
275 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
276 * Comment out request_contains_null_bytes() until it's used again.
278 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
279 * #1763173 from Stefan Huehner: s@const static@static const@
280 * and declare some more functions static.
282 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
283 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
284 * and explain why they were rejected in the first place.
285 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
286 * CONNECT requests. The request line was missing.
287 * - Add two more XXX reminders as we don't have enough already.
289 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
290 * As Hal noticed, checking dispatch_cgi() as the last cruncher
291 * looks like a bug if CGI requests are blocked unintentionally,
292 * so don't do it unless the user enabled the new config option
293 * "allow-cgi-request-crunching".
295 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
296 * - Allow to rewrite the request destination behind the client's back.
297 * - Turn the weird-looking unconditional for loop that
298 * reads the client request into a conditional while loop.
299 * Move the stuff that only runs once out of the loop.
300 * - Move parts of chat(), server_content_type() and the
301 * necessary stuff to fix BR#1750917 into get_filter_function().
303 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
304 * Temporary workaround for BR#1730105.
306 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
307 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
308 * deadlocks and crashes on OpenBSD and possibly other OS with neither
309 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
310 * Thanks to Ralf Horstmann for report and solution.
312 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
313 * Add forward-override{} to change the forwarding settings through
314 * action sections. This is mainly interesting to forward different
315 * clients differently (for example based on User-Agent or request
318 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
319 * - Let usage() mention the --chroot parameter.
320 * - Use read_socket() consistently and always leave
321 * the last buffer byte alone, even in cases where
322 * null termination (currently) doesn't matter.
324 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
325 * - Fix config file loading on Unix if no config file is specified.
326 * Since r1.97 Privoxy would always interpret the last argument as
327 * config file, even if it's a valid command line option.
328 * - Abort in case of unrecognized command line options. Closes #1719696.
329 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
330 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
332 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
333 * - Don't rerun crunchers that only depend on the request URL.
334 * - Don't count redirects and CGI requests as "blocked requests".
336 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
337 * Support crunching based on tags created by server-header taggers.
339 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
340 * Make HTTP snippets static (again). Add a Content-Type for those
341 * with content so the browser doesn't guess it based on the URL.
343 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
344 * Move crunching and request line rebuilding out of chat().
346 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
347 * Introduce tags as alternative way to specify which
348 * actions apply to a request. At the moment tags can be
349 * created based on client and server headers.
351 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
352 * Don't CLF-log CONNECT requests twice.
354 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
355 * Log the source address for ACL-related connection drops.
357 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
358 * New config option: enforce-blocks.
360 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
361 * - Move null byte check into separate function.
362 * - Don't confuse the client with error pages
363 * if a CONNECT request was already confirmed.
365 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
366 * Speed up NULL byte escaping and only log the complete
367 * NULL byte requests with header debugging enabled.
369 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
370 * Answer requests that contain NULL bytes with
371 * a custom response instead of waiting for more
372 * data until the client eventually hangs up.
374 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
375 * - Move delivery and logging of crunched responses
376 * from chat() into send_crunch_response().
377 * - Display the reason for generating http_responses.
378 * - Log the content length for LOG_LEVEL_CLF correctly
379 * (still incorrect for some fixed responses).
380 * - Reword an incorrect comment about
381 * treat-forbidden-connects-like-blocks violating
383 * - Add some log messages.
385 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
386 * Move mutex initialization into separate
387 * function and exit in case of errors.
389 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
390 * - Start to reduce chat()'s line count and move
391 * parts of it into separate functions.
392 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
394 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
395 * - Add Proxy-Agent header to HTTP snippets that are
396 * supposed to reach HTTP clients only.
397 * - Made a few CONNECT log messages more descriptive.
398 * - Catch completely empty server responses (as seen
399 * with Tor's fake ".noconnect" top level domain).
400 * - Use shiny new "forwarding-failed" template for socks errors.
402 * Revision 1.118 2007/01/07 07:43:43 joergs
403 * AmigaOS4 support added.
405 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
406 * Added config option accept-intercepted-requests
407 * and disabled it by default.
409 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
410 * Reverted parts of my last commit
411 * to keep error handling working.
413 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
414 * Fixed gcc43 conversion warnings.
416 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
417 * Fix -pedantic ISO C warning about converting
418 * from function pointer to object pointer.
420 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
421 * Silence compiler warning I introduced with my last commit.
423 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
424 * Mutex protect rand() if POSIX threading
425 * is used, warn the user if that's not possible
426 * and stop using it on _WIN32 where it could
429 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
430 * Don't prevent core dumps by catching SIGABRT.
431 * It's rude and makes debugging unreasonable painful.
433 * Revision 1.110 2006/12/13 14:52:53 etresoft
434 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
436 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
437 * Privoxy is now able to run as intercepting
438 * proxy in combination with any packet filter
439 * that does the port redirection. The destination
440 * is extracted from the "Host:" header which
441 * should be available for nearly all requests.
443 * Moved HTTP snipplets into jcc.c.
444 * Added error message for gopher proxy requests.
446 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
447 * Only unlink the pidfile if it's actually used.
449 * Change order of interception checks to make
450 * it possible to block or redirect requests for
453 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
454 * Make pthread mutex locking more generic. Instead of
455 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
456 * and use mutex locking unless there is an _r function
457 * available. Better safe than sorry.
459 * Fixes "./configure --disable-pthread" and should result
460 * in less threading-related problems on pthread-using platforms,
461 * but it still doesn't fix BR#1122404.
463 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
464 * Move pthread.h inclusion from jcc.c to jcc.h.
465 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
467 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
468 * Don't exit after receiving the second SIGHUP on Solaris.
470 * Fixes BR 1052235, but the same problem may exist on other
471 * systems. Once 3.0.6 is out we should use sigset()
472 * where available and see if it breaks anything.
474 * Revision 1.104 2006/09/23 13:26:38 roro
475 * Replace TABs by spaces in source code.
477 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
478 * Fix +redirect{}. Didn't work with -fast-redirects.
480 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
481 * Respond with 400 and a short text message
482 * if the client tries to use Privoxy as FTP proxy.
484 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
485 * Make number of retries in case of forwarded-connect problems
486 * a config file option (forwarded-connect-retries) and use 0 as
489 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
490 * Set random(3) seed.
492 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
493 * Follow the OpenBSD port's lead and protect the resolve
494 * functions on OpenBSD as well.
496 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
497 * --user fix. Only use the user as group if no group is specified.
498 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
500 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
501 * Windows service (re-)integration
505 * --install[:service_name]
506 * --uninstall[:service_name]
509 * They work as follows:
510 * --install will create a service for you and then terminate.
511 * By default the service name will be "privoxy" (without the quotes).
512 * However you can run multiple services if you wish, just by adding
513 * a colon and then a name (no spaces).
515 * --uninstall follows the exact same rules a --install.
517 * --service is used when the program is executed by the service
518 * control manager, and in normal circumstances would never be
519 * used as a command line argument.
521 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
522 * Windows service integration
524 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
525 * Incorporate Fabian Keil's patch work:
526 http://www.fabiankeil.de/sourcecode/privoxy/
528 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
529 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
530 * with what was really the latest development (the v_3_0_branch branch)
532 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
533 * Thanks to Jindrich Makovicka for a race condition fix for the log
534 * file. The race condition remains for non-pthread implementations.
535 * Reference patch #1175720.
537 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
538 * Add the ability to check jpeg images for invalid
539 * lengths of comment blocks. Defensive strategy
540 * against the exploit:
541 * Microsoft Security Bulletin MS04-028
542 * Buffer Overrun in JPEG Processing (GDI+) Could
543 * Allow Code Execution (833987)
544 * Enabled with +inspect-jpegs in actions files.
546 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
547 * - Fixed usage info for non-unix platforms
548 * - Fixed small cmdline parsing bug
550 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
551 * Diagnostics: Now preserve the returncode of pthread_create
552 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
554 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
555 * No longer ignore SIGCHLD. Fixes bug #769381
557 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
558 * Close jarfile on graceful exit, remove stray line
560 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
561 * Cosmetics: Killed a warning, a typo and an allocation left at exit
563 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
564 * No longer rely on non-POSIX.1 extensions of getcwd().
567 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
568 * Replaced setenv() by posix-compliant putenv()
569 * Thanks to Neil McCalden (nmcc AT users.sf.net).
571 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
572 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
574 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
575 * Clean-up and extension of improvements for forked mode:
576 * - Child's return code now consists of flags RC_FLAG_*
577 * - Reporting toggle to parent now properly #ifdef'ed
578 * - Children now report blocking to parent. This enables
579 * statistics in forked mode
581 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
582 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
583 * Children now report having been toggled through _exit(17), parents
584 * watch for that code and toggle themselves if found.
586 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
587 * Wrapping all *_r functions (the non-_r versions of them) with
588 * mutex semaphores for OSX. Hopefully this will take care of all
589 * of those pesky crash reports.
591 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
592 * Fixed two mostly harmless mem leaks
594 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
595 * Fix: Head of global clients list now initialized to NULL
597 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
598 * Added basic support for OPTIONS and TRACE HTTP methods:
599 * - New interceptor direct_response() added in chat().
600 * - sed() moved to earlier in the process, so that the
601 * Host: header is evaluated before actions and forwarding
604 * Revision 1.92 2002/05/08 16:00:46 oes
605 * Chat's buffer handling:
606 * - Fixed bug with unchecked out-of-mem conditions
607 * while reading client request & server headers
608 * - No longer predict if the buffer limit will be exceeded
609 * in the next read -- check add_to_iob's new
610 * return code. If buffer couldn't be extended
611 * (policy or out-of-mem) while
612 * - reading from client: abort
613 * - reading server headers: send error page
614 * - buffering server body for filter: flush,
615 * and if that fails: send error page
617 * Revision 1.91 2002/04/08 20:35:58 swa
620 * Revision 1.90 2002/04/02 14:57:28 oes
621 * Made sending wafers independent of FEATURE_COOKIE_JAR
623 * Revision 1.89 2002/03/31 17:18:59 jongfoster
624 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
626 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
627 * More compiler warning message maintenance
629 * Revision 1.87 2002/03/26 22:29:54 swa
630 * we have a new homepage!
632 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
633 * Workaround for closing the jarfile before load_config() comes around again
635 * Revision 1.85 2002/03/24 15:23:33 jongfoster
638 * Revision 1.84 2002/03/24 13:25:43 swa
639 * name change related issues
641 * Revision 1.83 2002/03/16 23:54:06 jongfoster
642 * Adding graceful termination feature, to help look for memory leaks.
643 * If you enable this (which, by design, has to be done by hand
644 * editing config.h) and then go to http://i.j.b/die, then the program
645 * will exit cleanly after the *next* request. It should free all the
646 * memory that was used.
648 * Revision 1.82 2002/03/13 00:27:05 jongfoster
651 * Revision 1.81 2002/03/12 01:42:50 oes
652 * Introduced modular filters
654 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
655 * OS/2 port maintenance:
656 * - Fixed EMX build - it had decayed a little
657 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
658 * substituted a memset for now.
660 * Revision 1.79 2002/03/09 20:03:52 jongfoster
661 * - Making various functions return int rather than size_t.
662 * (Undoing a recent change). Since size_t is unsigned on
663 * Windows, functions like read_socket that return -1 on
664 * error cannot return a size_t.
666 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
667 * crashes, and also frequently caused JB to jump to 100%
668 * CPU and stay there. (Because it thought it had just
669 * read ((unsigned)-1) == 4Gb of data...)
671 * - The signature of write_socket has changed, it now simply
672 * returns success=0/failure=nonzero.
674 * - Trying to get rid of a few warnings --with-debug on
675 * Windows, I've introduced a new type "jb_socket". This is
676 * used for the socket file descriptors. On Windows, this
677 * is SOCKET (a typedef for unsigned). Everywhere else, it's
678 * an int. The error value can't be -1 any more, so it's
679 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
680 * Windows it maps to the #define INVALID_SOCKET.)
682 * - The signature of bind_port has changed.
684 * Revision 1.78 2002/03/08 21:35:04 oes
685 * Added optional group supplement to --user option. Will now use default group of user if no group given
687 * Revision 1.77 2002/03/07 03:52:06 oes
688 * - Fixed compiler warnings etc
689 * - Improved handling of failed DNS lookups
691 * Revision 1.76 2002/03/06 22:54:35 jongfoster
692 * Automated function-comment nitpicking.
694 * Revision 1.75 2002/03/06 10:02:19 oes
695 * Fixed stupid bug when --user was not given
697 * Revision 1.74 2002/03/06 00:49:31 jongfoster
698 * Fixing warning on Windows
699 * Making #ifdefs that refer to the same variable consistently
700 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
702 * Revision 1.73 2002/03/05 23:57:30 hal9
703 * Stray character 's' on line 1618 was breaking build.
705 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
706 * - Re-enable OS/2 building after new parms were added
707 * - Fix false out of memory report when resolving CGI templates when no IP
708 * address is available of failed attempt (a la no such domain)
710 * Revision 1.71 2002/03/05 18:13:56 oes
711 * Added --user option
713 * Revision 1.70 2002/03/05 04:52:42 oes
714 * Deleted non-errlog debugging code
716 * Revision 1.69 2002/03/04 23:50:00 jongfoster
717 * Splitting off bind_port() call into bind_port_helper(), with
720 * Revision 1.68 2002/03/04 20:17:32 oes
723 * Revision 1.67 2002/03/04 18:18:57 oes
724 * - Removed _DEBUG mode
725 * - Cleand up cmdline parsing
726 * - Introduced --no-daemon, --pidfile options
727 * - Cleaned up signal handling:
728 * - Terminate cleanly on INT, TERM and ABRT
729 * - Schedule logfile for re-opening on HUP
730 * - Ignore CHLD and PIPE
731 * - Leave the rest with their default handlers
732 * - Uniform handler registration
733 * - Added usage() function
734 * - Played styleguide police
736 * Revision 1.66 2002/03/03 15:06:55 oes
737 * Re-enabled automatic config reloading
739 * Revision 1.65 2002/03/03 14:49:11 oes
740 * Fixed CLF logging: Now uses client's original HTTP request
742 * Revision 1.64 2002/03/03 09:18:03 joergs
743 * Made jumbjuster work on AmigaOS again.
745 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
746 * Clean up a little CRLF unpleasantness that suddenly appeared
748 * Revision 1.62 2002/02/20 23:17:23 jongfoster
749 * Detecting some out-of memory conditions and exiting with a log message.
751 * Revision 1.61 2002/01/17 21:01:52 jongfoster
752 * Moving all our URL and URL pattern parsing code to urlmatch.c.
754 * Revision 1.60 2001/12/30 14:07:32 steudten
755 * - Add signal handling (unix)
756 * - Add SIGHUP handler (unix)
757 * - Add creation of pidfile (unix)
758 * - Add action 'top' in rc file (RH)
759 * - Add entry 'SIGNALS' to manpage
760 * - Add exit message to logfile (unix)
762 * Revision 1.59 2001/12/13 14:07:18 oes
763 * Fixed Bug: 503 error page now sent OK
765 * Revision 1.58 2001/11/30 23:37:24 jongfoster
766 * Renaming the Win32 config file to config.txt - this is almost the
767 * same as the corresponding UNIX name "config"
769 * Revision 1.57 2001/11/16 00:47:43 jongfoster
770 * Changing the tty-disconnection code to use setsid().
772 * Revision 1.56 2001/11/13 20:20:54 jongfoster
773 * Tabs->spaces, fixing a bug with missing {} around an if()
775 * Revision 1.55 2001/11/13 20:14:53 jongfoster
776 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
778 * Revision 1.54 2001/11/07 00:03:14 steudten
779 * Give reliable return value if an error
780 * occurs not just 0 with new daemon mode.
782 * Revision 1.53 2001/11/05 21:41:43 steudten
783 * Add changes to be a real daemon just for unix os.
784 * (change cwd to /, detach from controlling tty, set
785 * process group and session leader to the own process.
787 * Add some fatal-error log message for failed malloc().
788 * Add '-d' if compiled with 'configure --with-debug' to
789 * enable debug output.
791 * Revision 1.52 2001/10/26 20:11:20 jongfoster
792 * Fixing type mismatch
794 * Revision 1.51 2001/10/26 17:38:28 oes
797 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
798 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
799 * threads to call select() simultaneously. So, it's time to do a real, live,
800 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
801 * (native). Both versions will work, but using __OS2__ offers multi-threading.
803 * Revision 1.49 2001/10/23 21:41:35 jongfoster
804 * Added call to initialize the (statically-allocated of course)
805 * "out of memory" CGI response.
807 * Revision 1.48 2001/10/10 19:56:46 jongfoster
808 * Moving some code that wasn't cookie-related out of an #ifdef
811 * Revision 1.47 2001/10/10 16:44:36 oes
812 * Added CONNECT destination port limitation check
814 * Revision 1.46 2001/10/08 15:17:41 oes
815 * Re-enabled SSL forwarding
817 * Revision 1.45 2001/10/07 15:42:11 oes
818 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
820 * Moved downgrading of the HTTP version from parse_http_request to
821 * chat(), since we can't decide if it is necessary before we have
822 * determined the actions for the URL. The HTTP command is now
823 * *always* re-built so the repairs need no longer be special-cased.
825 * filter_popups now gets a csp pointer so it can raise the new
826 * CSP_FLAG_MODIFIED flag.
830 * Added configurable size limit for the IOB. If the IOB grows so
831 * large that the next read would exceed the limit, the header
832 * is generated, and the header & unfiltered buffer are flushed
833 * to the client. Chat then continues in non-buffering,
834 * non-filtering body mode.
836 * Revision 1.44 2001/10/02 18:13:57 oes
839 * Revision 1.43 2001/10/02 15:32:13 oes
840 * Moved generation of hdr
842 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
843 * Cleaning up 2 compiler warnings on OS/2.
845 * Revision 1.41 2001/09/16 17:05:14 jongfoster
846 * Removing unused #include showarg.h
848 * Revision 1.40 2001/09/16 15:41:45 jongfoster
849 * Fixing signed/unsigned comparison warning.
851 * Revision 1.39 2001/09/16 13:21:27 jongfoster
852 * Changes to use new list functions.
854 * Revision 1.38 2001/09/16 13:01:46 jongfoster
855 * Removing redundant function call that zeroed zalloc()'d memory.
857 * Revision 1.37 2001/09/10 11:12:24 oes
858 * Deleted unused variable
860 * Revision 1.36 2001/09/10 10:56:15 oes
861 * Silenced compiler warnings
863 * Revision 1.35 2001/07/31 14:44:22 oes
864 * Deleted unused size parameter from filter_popups()
866 * Revision 1.34 2001/07/30 22:08:36 jongfoster
867 * Tidying up #defines:
868 * - All feature #defines are now of the form FEATURE_xxx
869 * - Permanently turned off WIN_GUI_EDIT
870 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
872 * Revision 1.33 2001/07/29 19:32:00 jongfoster
873 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
875 * Revision 1.32 2001/07/29 18:47:05 jongfoster
876 * Adding missing #include "loadcfg.h"
878 * Revision 1.31 2001/07/29 12:17:48 oes
879 * Applied pthread fix by Paul Lieverse
881 * Revision 1.30 2001/07/25 22:57:13 jongfoster
882 * __BEOS__ no longer overrides FEATURE_PTHREAD.
883 * This is because FEATURE_PTHREAD will soon be widely used, so I
884 * want to keep it simple.
886 * Revision 1.29 2001/07/24 12:47:06 oes
887 * Applied BeOS support update by Eugenia
889 * Revision 1.28 2001/07/23 13:26:12 oes
890 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
892 * Revision 1.27 2001/07/19 19:09:47 haroon
893 * - Added code to take care of the situation where while processing the first
894 * server response (which includes the server header), after finding the end
895 * of the headers we were not looking past the end of the headers for
896 * content modification. I enabled it for filter_popups.
897 * Someone else should look to see if other similar operations should be
898 * done to the discarded portion of the buffer.
900 * Note 2001/07/20: No, the other content modification mechanisms will process
901 * the whole iob later anyway. --oes
903 * Revision 1.26 2001/07/18 12:31:36 oes
906 * Revision 1.25 2001/07/15 19:43:49 jongfoster
907 * Supports POSIX threads.
908 * Also removed some unused #includes.
910 * Revision 1.24 2001/07/13 14:00:40 oes
911 * - Generic content modification scheme:
912 * Each feature has its own applicability flag that is set
913 * from csp->action->flags.
914 * Replaced the "filtering" int flag , by a function pointer
915 * "content_filter" to the function that will do the content
916 * modification. If it is != NULL, the document will be buffered
917 * and processed through *content_filter, which must set
918 * csp->content_length and return a modified copy of the body
919 * or return NULL (on failiure).
920 * - Changed csp->is_text to the more generic bitmap csp->content_type
921 * which can currently take the valued CT_TEXT or CT_GIF
923 * - Removed all #ifdef PCRS
925 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
926 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
927 * the `block_acl' call. This prevents a compilation error when the user
928 * does not wish to use the "ACL" feature.
930 * Revision 1.22 2001/06/29 21:45:41 oes
931 * Indentation, CRLF->LF, Tab-> Space
933 * Revision 1.21 2001/06/29 13:29:36 oes
934 * - Cleaned up, improved comments
935 * - Unified all possible interceptors (CGI,
936 * block, trust, fast_redirect) in one
937 * place, with one (CGI) answer generation
938 * mechansim. Much clearer now.
939 * - Removed the GIF image generation, which
940 * is now done in filters.c:block_url()
941 * - Made error conditions like domain lookup
942 * failiure or (various) problems while talking
943 * to the server use cgi.c:error_response()
944 * instead of generating HTML/HTTP in chat() (yuck!)
945 * - Removed logentry from cancelled commit
947 * Revision 1.20 2001/06/09 10:55:28 jongfoster
948 * Changing BUFSIZ ==> BUFFER_SIZE
950 * Revision 1.19 2001/06/07 23:12:52 jongfoster
951 * Replacing function pointer in struct gateway with a directly
952 * called function forwarded_connect().
953 * Replacing struct gateway with struct forward_spec
955 * Revision 1.18 2001/06/03 19:12:16 oes
956 * introduced new cgi handling
958 * Revision 1.17 2001/06/01 20:07:23 jongfoster
959 * Now uses action +image-blocker{} rather than config->tinygif
961 * Revision 1.16 2001/06/01 18:49:17 jongfoster
962 * Replaced "list_share" with "list" - the tiny memory gain was not
963 * worth the extra complexity.
965 * Revision 1.15 2001/05/31 21:24:47 jongfoster
966 * Changed "permission" to "action" throughout.
967 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
968 * Moved vanilla wafer check into chat(), since we must now
969 * decide whether or not to add it based on the URL.
971 * Revision 1.14 2001/05/29 20:14:01 joergs
972 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
975 * Revision 1.13 2001/05/29 09:50:24 jongfoster
976 * Unified blocklist/imagelist/permissionslist.
977 * File format is still under discussion, but the internal changes
980 * Also modified interceptor behaviour:
981 * - We now intercept all URLs beginning with one of the following
982 * prefixes (and *only* these prefixes):
984 * * http://ijbswa.sf.net/config/
985 * * http://ijbswa.sourceforge.net/config/
986 * - New interceptors "home page" - go to http://i.j.b/ to see it.
987 * - Internal changes so that intercepted and fast redirect pages
988 * are not replaced with an image.
989 * - Interceptors now have the option to send a binary page direct
990 * to the client. (i.e. ijb-send-banner uses this)
991 * - Implemented show-url-info interceptor. (Which is why I needed
992 * the above interceptors changes - a typical URL is
993 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
994 * The previous mechanism would not have intercepted that, and
995 * if it had been intercepted then it then it would have replaced
998 * Revision 1.12 2001/05/27 22:17:04 oes
1000 * - re_process_buffer no longer writes the modified buffer
1001 * to the client, which was very ugly. It now returns the
1002 * buffer, which it is then written by chat.
1004 * - content_length now adjusts the Content-Length: header
1005 * for modified documents rather than crunch()ing it.
1006 * (Length info in csp->content_length, which is 0 for
1007 * unmodified documents)
1009 * - For this to work, sed() is called twice when filtering.
1011 * Revision 1.11 2001/05/26 17:27:53 jongfoster
1012 * Added support for CLF and fixed LOG_LEVEL_LOG.
1013 * Also did CRLF->LF fix of my previous patch.
1015 * Revision 1.10 2001/05/26 15:26:15 jongfoster
1016 * ACL feature now provides more security by immediately dropping
1017 * connections from untrusted hosts.
1019 * Revision 1.9 2001/05/26 00:28:36 jongfoster
1020 * Automatic reloading of config file.
1021 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
1022 * Most of the global variables have been moved to a new
1023 * struct configuration_spec, accessed through csp->config->globalname
1024 * Most of the globals remaining are used by the Win32 GUI.
1026 * Revision 1.8 2001/05/25 22:43:18 jongfoster
1027 * Fixing minor memory leak and buffer overflow.
1029 * Revision 1.7 2001/05/25 22:34:30 jongfoster
1032 * Revision 1.6 2001/05/23 00:13:58 joergs
1033 * AmigaOS support fixed.
1035 * Revision 1.5 2001/05/22 18:46:04 oes
1037 * - Enabled filtering banners by size rather than URL
1038 * by adding patterns that replace all standard banner
1039 * sizes with the "Junkbuster" gif to the re_filterfile
1041 * - Enabled filtering WebBugs by providing a pattern
1042 * which kills all 1x1 images
1044 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
1045 * which is selected by the (nonstandard and therefore
1046 * capital) letter 'U' in the option string.
1047 * It causes the quantifiers to be ungreedy by default.
1048 * Appending a ? turns back to greedy (!).
1050 * - Added a new interceptor ijb-send-banner, which
1051 * sends back the "Junkbuster" gif. Without imagelist or
1052 * MSIE detection support, or if tinygif = 1, or the
1053 * URL isn't recognized as an imageurl, a lame HTML
1054 * explanation is sent instead.
1056 * - Added new feature, which permits blocking remote
1057 * script redirects and firing back a local redirect
1059 * The feature is conditionally compiled, i.e. it
1060 * can be disabled with --disable-fast-redirects,
1061 * plus it must be activated by a "fast-redirects"
1062 * line in the config file, has its own log level
1063 * and of course wants to be displayed by show-proxy-args
1064 * Note: Boy, all the #ifdefs in 1001 locations and
1065 * all the fumbling with configure.in and acconfig.h
1066 * were *way* more work than the feature itself :-(
1068 * - Because a generic redirect template was needed for
1069 * this, tinygif = 3 now uses the same.
1071 * - Moved GIFs, and other static HTTP response templates
1074 * - Some minor fixes
1076 * - Removed some >400 CRs again (Jon, you really worked
1079 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1080 * Made failure to bind() a fatal error.
1082 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1083 * Version 2.9.4 checkin.
1084 * - Merged popupfile and cookiefile, and added control over PCRS
1085 * filtering, in new "permissionsfile".
1086 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1087 * file error you now get a message box (in the Win32 GUI) rather
1088 * than the program exiting with no explanation.
1089 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1091 * - Removed tabs from "config"
1092 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1093 * - Bumped up version number.
1095 * Revision 1.2 2001/05/17 22:34:44 oes
1096 * - Added hint on GIF char array generation to jcc.c
1097 * - Cleaned CRLF's from the sources and related files
1098 * - Repaired logging for REF and FRC
1100 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1101 * Initial import of version 2.9.3 source tree
1104 *********************************************************************/
1110 #include <sys/types.h>
1119 # ifndef FEATURE_PTHREAD
1123 # include <windows.h>
1124 # include <process.h>
1125 # endif /* ndef FEATURE_PTHREAD */
1128 # ifndef _WIN_CONSOLE
1129 # include "w32log.h"
1130 # endif /* ndef _WIN_CONSOLE */
1131 # include "w32svrapi.h"
1133 #else /* ifndef _WIN32 */
1135 # if !defined (__OS2__)
1136 # include <unistd.h>
1137 # include <sys/wait.h>
1138 # endif /* ndef __OS2__ */
1139 # include <sys/time.h>
1140 # include <sys/stat.h>
1141 # include <sys/ioctl.h>
1144 #include <sys/termios.h>
1152 # include <signal.h>
1155 # include <socket.h> /* BeOS has select() for sockets only. */
1156 # include <OS.h> /* declarations for threads and stuff. */
1159 # if defined(__EMX__) || defined(__OS2__)
1160 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1165 #define bzero(B,N) memset(B,0x00,n)
1169 # include <select.h>
1174 #include "project.h"
1177 #include "filters.h"
1178 #include "loaders.h"
1179 #include "parsers.h"
1180 #include "miscutil.h"
1182 #include "jbsockets.h"
1183 #include "gateway.h"
1184 #include "actions.h"
1186 #include "loadcfg.h"
1187 #include "urlmatch.h"
1189 const char jcc_h_rcs[] = JCC_H_VERSION;
1190 const char project_h_rcs[] = PROJECT_H_VERSION;
1193 struct client_state clients[1];
1194 struct file_list files[1];
1196 #ifdef FEATURE_STATISTICS
1197 int urls_read = 0; /* total nr of urls read inc rejected */
1198 int urls_rejected = 0; /* total nr of urls rejected */
1199 #endif /* def FEATURE_STATISTICS */
1201 #ifdef FEATURE_GRACEFUL_TERMINATION
1202 int g_terminate = 0;
1205 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1206 static void sig_handler(int the_signal);
1208 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1209 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1210 static jb_err get_server_headers(struct client_state *csp);
1211 static const char *crunch_reason(const struct http_response *rsp);
1212 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1213 static char *get_request_line(struct client_state *csp);
1214 static jb_err receive_client_request(struct client_state *csp);
1215 static jb_err parse_client_request(struct client_state *csp);
1216 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1217 static jb_err change_request_destination(struct client_state *csp);
1218 static void chat(struct client_state *csp);
1219 static void serve(struct client_state *csp);
1220 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1221 static void usage(const char *myname);
1223 static void initialize_mutexes(void);
1224 static jb_socket bind_port_helper(struct configuration_spec *config);
1225 static void listen_loop(void);
1228 void serve(struct client_state *csp);
1229 #else /* ifndef AMIGA */
1230 static void serve(struct client_state *csp);
1231 #endif /* def AMIGA */
1234 static int32 server_thread(void *data);
1235 #endif /* def __BEOS__ */
1238 #define sleep(N) Sleep(((N) * 1000))
1242 #define sleep(N) DosSleep(((N) * 100))
1245 #ifdef MUTEX_LOCKS_AVAILABLE
1247 * XXX: Does the locking stuff really belong in this file?
1249 privoxy_mutex_t log_mutex;
1250 privoxy_mutex_t log_init_mutex;
1251 privoxy_mutex_t connection_reuse_mutex;
1253 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1254 privoxy_mutex_t resolver_mutex;
1255 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1257 #ifndef HAVE_GMTIME_R
1258 privoxy_mutex_t gmtime_mutex;
1259 #endif /* ndef HAVE_GMTIME_R */
1261 #ifndef HAVE_LOCALTIME_R
1262 privoxy_mutex_t localtime_mutex;
1263 #endif /* ndef HAVE_GMTIME_R */
1266 privoxy_mutex_t rand_mutex;
1267 #endif /* ndef HAVE_RANDOM */
1269 #endif /* def MUTEX_LOCKS_AVAILABLE */
1272 const char *basedir = NULL;
1273 const char *pidfile = NULL;
1274 static int received_hup_signal = 0;
1275 #endif /* defined unix */
1277 /* HTTP snipplets. */
1278 static const char CSUCCEED[] =
1279 "HTTP/1.0 200 Connection established\r\n"
1280 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1282 static const char CHEADER[] =
1283 "HTTP/1.0 400 Invalid header received from client\r\n"
1284 "Proxy-Agent: Privoxy " VERSION "\r\n"
1285 "Content-Type: text/plain\r\n"
1286 "Connection: close\r\n\r\n"
1287 "Invalid header received from client.\r\n";
1289 static const char FTP_RESPONSE[] =
1290 "HTTP/1.0 400 Invalid request received from client\r\n"
1291 "Content-Type: text/plain\r\n"
1292 "Connection: close\r\n\r\n"
1293 "Invalid request. Privoxy doesn't support FTP.\r\n";
1295 static const char GOPHER_RESPONSE[] =
1296 "HTTP/1.0 400 Invalid request received from client\r\n"
1297 "Content-Type: text/plain\r\n"
1298 "Connection: close\r\n\r\n"
1299 "Invalid request. Privoxy doesn't support gopher.\r\n";
1301 /* XXX: should be a template */
1302 static const char MISSING_DESTINATION_RESPONSE[] =
1303 "HTTP/1.0 400 Bad request received from client\r\n"
1304 "Proxy-Agent: Privoxy " VERSION "\r\n"
1305 "Content-Type: text/plain\r\n"
1306 "Connection: close\r\n\r\n"
1307 "Bad request. Privoxy was unable to extract the destination.\r\n";
1309 /* XXX: should be a template */
1310 static const char NO_SERVER_DATA_RESPONSE[] =
1311 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1312 "Proxy-Agent: Privoxy " VERSION "\r\n"
1313 "Content-Type: text/plain\r\n"
1314 "Connection: close\r\n\r\n"
1315 "Empty server or forwarder response.\r\n"
1316 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1318 /* XXX: should be a template */
1319 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1320 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1321 "Proxy-Agent: Privoxy " VERSION "\r\n"
1322 "Content-Type: text/plain\r\n"
1323 "Connection: close\r\n\r\n"
1324 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1327 /* XXX: should be a template */
1328 static const char NULL_BYTE_RESPONSE[] =
1329 "HTTP/1.0 400 Bad request received from client\r\n"
1330 "Proxy-Agent: Privoxy " VERSION "\r\n"
1331 "Content-Type: text/plain\r\n"
1332 "Connection: close\r\n\r\n"
1333 "Bad request. Null byte(s) before end of request.\r\n";
1336 /* XXX: should be a template */
1337 static const char MESSED_UP_REQUEST_RESPONSE[] =
1338 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1339 "Proxy-Agent: Privoxy " VERSION "\r\n"
1340 "Content-Type: text/plain\r\n"
1341 "Connection: close\r\n\r\n"
1342 "Bad request. Messed up with header filters.\r\n";
1344 /* A function to crunch a response */
1345 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1347 /* Crunch function flags */
1348 #define CF_NO_FLAGS 0
1349 /* Cruncher applies to forced requests as well */
1350 #define CF_IGNORE_FORCE 1
1351 /* Crunched requests are counted for the block statistics */
1352 #define CF_COUNT_AS_REJECT 2
1354 /* A crunch function and its flags */
1357 const crunch_func_ptr cruncher;
1361 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1363 /* Complete list of cruncher functions */
1364 static const struct cruncher crunchers_all[] = {
1365 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1366 { block_url, CF_COUNT_AS_REJECT },
1367 #ifdef FEATURE_TRUST
1368 { trust_url, CF_COUNT_AS_REJECT },
1369 #endif /* def FEATURE_TRUST */
1370 { redirect_url, CF_NO_FLAGS },
1371 { dispatch_cgi, CF_IGNORE_FORCE},
1375 /* Light version, used after tags are applied */
1376 static const struct cruncher crunchers_light[] = {
1377 { block_url, CF_COUNT_AS_REJECT },
1378 { redirect_url, CF_NO_FLAGS },
1384 * XXX: Don't we really mean
1390 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1391 /*********************************************************************
1393 * Function : sig_handler
1395 * Description : Signal handler for different signals.
1396 * Exit gracefully on TERM and INT
1397 * or set a flag that will cause the errlog
1398 * to be reopened by the main thread on HUP.
1401 * 1 : the_signal = the signal cause this function to call
1405 *********************************************************************/
1406 static void sig_handler(int the_signal)
1412 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1424 received_hup_signal = 1;
1430 * We shouldn't be here, unless we catch signals
1431 * in main() that we can't handle here!
1433 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1441 /*********************************************************************
1443 * Function : client_protocol_is_unsupported
1445 * Description : Checks if the client used a known unsupported
1446 * protocol and deals with it by sending an error
1450 * 1 : csp = Current client state (buffers, headers, etc...)
1451 * 2 : req = the first request line send by the client
1453 * Returns : TRUE if an error response has been generated, or
1454 * FALSE if the request doesn't look invalid.
1456 *********************************************************************/
1457 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1460 * If it's a FTP or gopher request, we don't support it.
1462 * These checks are better than nothing, but they might
1463 * not work in all configurations and some clients might
1464 * have problems digesting the answer.
1466 * They should, however, never cause more problems than
1467 * Privoxy's old behaviour (returning the misleading HTML
1470 * "Could not resolve http://(ftp|gopher)://example.org").
1472 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1474 const char *response = NULL;
1475 const char *protocol = NULL;
1477 if (!strncmpic(req, "GET ftp://", 10))
1479 response = FTP_RESPONSE;
1484 response = GOPHER_RESPONSE;
1485 protocol = "GOPHER";
1487 log_error(LOG_LEVEL_ERROR,
1488 "%s tried to use Privoxy as %s proxy: %s",
1489 csp->ip_addr_str, protocol, req);
1490 log_error(LOG_LEVEL_CLF,
1491 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1493 write_socket(csp->cfd, response, strlen(response));
1502 /*********************************************************************
1504 * Function : get_request_destination_elsewhere
1506 * Description : If the client's request was redirected into
1507 * Privoxy without the client's knowledge,
1508 * the request line lacks the destination host.
1510 * This function tries to get it elsewhere,
1511 * provided accept-intercepted-requests is enabled.
1513 * "Elsewhere" currently only means "Host: header",
1514 * but in the future we may ask the redirecting
1515 * packet filter to look the destination up.
1517 * If the destination stays unknown, an error
1518 * response is send to the client and headers
1519 * are freed so that chat() can return directly.
1522 * 1 : csp = Current client state (buffers, headers, etc...)
1523 * 2 : headers = a header list
1525 * Returns : JB_ERR_OK if the destination is now known, or
1526 * JB_ERR_PARSE if it isn't.
1528 *********************************************************************/
1529 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1533 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1535 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1536 " Privoxy isn't configured to accept intercepted requests.",
1537 csp->ip_addr_str, csp->http->cmd);
1538 /* XXX: Use correct size */
1539 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1540 csp->ip_addr_str, csp->http->cmd);
1542 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1543 destroy_list(headers);
1545 return JB_ERR_PARSE;
1547 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1549 /* Split the domain we just got for pattern matching */
1550 init_domain_components(csp->http);
1556 /* We can't work without destination. Go spread the news.*/
1558 req = list_to_text(headers);
1560 /* XXX: Use correct size */
1561 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1562 csp->ip_addr_str, csp->http->cmd);
1563 log_error(LOG_LEVEL_ERROR,
1564 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1565 csp->ip_addr_str, csp->http->cmd, req);
1568 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1569 destroy_list(headers);
1571 return JB_ERR_PARSE;
1574 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1575 * to get the destination IP address, use it as host directly
1576 * or do a reverse DNS lookup first.
1581 /*********************************************************************
1583 * Function : get_server_headers
1585 * Description : Parses server headers in iob and fills them
1586 * into csp->headers so that they can later be
1590 * 1 : csp = Current client state (buffers, headers, etc...)
1592 * Returns : JB_ERR_OK if everything went fine, or
1593 * JB_ERR_PARSE if the headers were incomplete.
1595 *********************************************************************/
1596 static jb_err get_server_headers(struct client_state *csp)
1598 int continue_hack_in_da_house = 0;
1601 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1606 * continue hack in da house. Ignore the ending of
1607 * this head and continue enlisting header lines.
1608 * The reason is described below.
1610 enlist(csp->headers, "");
1611 continue_hack_in_da_house = 0;
1614 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1617 * It's a bodyless continue response, don't
1618 * stop header parsing after reaching its end.
1620 * As a result Privoxy will concatenate the
1621 * next response's head and parse and deliver
1622 * the headers as if they belonged to one request.
1624 * The client will separate them because of the
1625 * empty line between them.
1627 * XXX: What we're doing here is clearly against
1628 * the intended purpose of the continue header,
1629 * and under some conditions (HTTP/1.0 client request)
1630 * it's a standard violation.
1632 * Anyway, "sort of against the spec" is preferable
1633 * to "always getting confused by Continue responses"
1634 * (Privoxy's behaviour before this hack was added)
1636 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1637 continue_hack_in_da_house = 1;
1639 else if (*header == '\0')
1642 * If the header is empty, but the Continue hack
1643 * isn't active, we can assume that we reached the
1644 * end of the buffer before we hit the end of the
1647 * Inform the caller an let it decide how to handle it.
1649 return JB_ERR_PARSE;
1652 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1655 * XXX: Should we quit the request and return a
1656 * out of memory error page instead?
1658 log_error(LOG_LEVEL_ERROR,
1659 "Out of memory while enlisting server headers. %s lost.",
1669 /*********************************************************************
1671 * Function : crunch_reason
1673 * Description : Translates the crunch reason code into a string.
1676 * 1 : rsp = a http_response
1678 * Returns : A string with the crunch reason or an error description.
1680 *********************************************************************/
1681 static const char *crunch_reason(const struct http_response *rsp)
1683 char * reason = NULL;
1685 assert(rsp != NULL);
1688 return "Internal error while searching for crunch reason";
1691 switch (rsp->reason)
1693 case RSP_REASON_UNSUPPORTED:
1694 reason = "Unsupported HTTP feature";
1696 case RSP_REASON_BLOCKED:
1699 case RSP_REASON_UNTRUSTED:
1700 reason = "Untrusted";
1702 case RSP_REASON_REDIRECTED:
1703 reason = "Redirected";
1705 case RSP_REASON_CGI_CALL:
1706 reason = "CGI Call";
1708 case RSP_REASON_NO_SUCH_DOMAIN:
1709 reason = "DNS failure";
1711 case RSP_REASON_FORWARDING_FAILED:
1712 reason = "Forwarding failed";
1714 case RSP_REASON_CONNECT_FAILED:
1715 reason = "Connection failure";
1717 case RSP_REASON_OUT_OF_MEMORY:
1718 reason = "Out of memory (may mask other reasons)";
1721 reason = "No reason recorded";
1729 /*********************************************************************
1731 * Function : send_crunch_response
1733 * Description : Delivers already prepared response for
1734 * intercepted requests, logs the interception
1735 * and frees the response.
1738 * 1 : csp = Current client state (buffers, headers, etc...)
1739 * 1 : rsp = Fully prepared response. Will be freed on exit.
1741 * Returns : Nothing.
1743 *********************************************************************/
1744 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1746 const struct http_request *http = csp->http;
1747 char status_code[4];
1749 assert(rsp != NULL);
1750 assert(rsp->head != NULL);
1755 * Not supposed to happen. If it does
1756 * anyway, treat it as an unknown error.
1758 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1759 /* return code doesn't matter */
1764 /* If rsp is still NULL, we have serious internal problems. */
1765 log_error(LOG_LEVEL_FATAL,
1766 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1770 * Extract the status code from the actual head
1771 * that was send to the client. It is the only
1772 * way to get it right for all requests, including
1773 * the fixed ones for out-of-memory problems.
1775 * A head starts like this: 'HTTP/1.1 200...'
1779 status_code[0] = rsp->head[9];
1780 status_code[1] = rsp->head[10];
1781 status_code[2] = rsp->head[11];
1782 status_code[3] = '\0';
1784 /* Write the answer to the client */
1785 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1786 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1788 /* There is nothing we can do about it. */
1789 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1792 /* Log that the request was crunched and why. */
1793 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1794 http->hostport, http->path, crunch_reason(rsp));
1795 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1796 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1798 /* Clean up and return */
1799 if (cgi_error_memory() != rsp)
1801 free_http_response(rsp);
1808 /*********************************************************************
1810 * Function : request_contains_null_bytes
1812 * Description : Checks for NULL bytes in the request and sends
1813 * an error message to the client if any were found.
1815 * XXX: currently not used, see comment in chat().
1818 * 1 : csp = Current client state (buffers, headers, etc...)
1819 * 2 : buf = Data from the client's request to check.
1820 * 3 : len = The data length.
1822 * Returns : TRUE if the request contained one or more NULL bytes, or
1825 *********************************************************************/
1826 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1828 size_t c_len; /* Request lenght when treated as C string */
1830 c_len = strlen(buf);
1835 * Null byte(s) found. Log the request,
1836 * return an error response and hang up.
1838 size_t tmp_len = c_len;
1843 * Replace NULL byte(s) with '°' characters
1844 * so the request can be logged as string.
1845 * XXX: Is there a better replacement character?
1848 tmp_len += strlen(buf+tmp_len);
1849 } while (tmp_len < len);
1851 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1852 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1853 log_error(LOG_LEVEL_HEADER,
1854 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1856 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1858 /* XXX: Log correct size */
1859 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1869 /*********************************************************************
1871 * Function : crunch_response_triggered
1873 * Description : Checks if the request has to be crunched,
1874 * and delivers the crunch response if necessary.
1877 * 1 : csp = Current client state (buffers, headers, etc...)
1878 * 2 : crunchers = list of cruncher functions to run
1880 * Returns : TRUE if the request was answered with a crunch response
1883 *********************************************************************/
1884 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1886 struct http_response *rsp = NULL;
1887 const struct cruncher *c;
1890 * If CGI request crunching is disabled,
1891 * check the CGI dispatcher out of order to
1892 * prevent unintentional blocks or redirects.
1894 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1895 && (NULL != (rsp = dispatch_cgi(csp))))
1897 /* Deliver, log and free the interception response. */
1898 send_crunch_response(csp, rsp);
1902 for (c = crunchers; c->cruncher != NULL; c++)
1905 * Check the cruncher if either Privoxy is toggled
1906 * on and the request isn't forced, or if the cruncher
1907 * applies to forced requests as well.
1909 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1910 !(csp->flags & CSP_FLAG_FORCED)) ||
1911 (c->flags & CF_IGNORE_FORCE))
1913 rsp = c->cruncher(csp);
1916 /* Deliver, log and free the interception response. */
1917 send_crunch_response(csp, rsp);
1918 #ifdef FEATURE_STATISTICS
1919 if (c->flags & CF_COUNT_AS_REJECT)
1921 csp->flags |= CSP_FLAG_REJECTED;
1923 #endif /* def FEATURE_STATISTICS */
1934 /*********************************************************************
1936 * Function : build_request_line
1938 * Description : Builds the HTTP request line.
1940 * If a HTTP forwarder is used it expects the whole URL,
1941 * web servers only get the path.
1944 * 1 : csp = Current client state (buffers, headers, etc...)
1945 * 2 : fwd = The forwarding spec used for the request
1946 * XXX: Should use http->fwd instead.
1947 * 3 : request_line = The old request line which will be replaced.
1949 * Returns : Nothing. Terminates in case of memory problems.
1951 *********************************************************************/
1952 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1954 struct http_request *http = csp->http;
1956 assert(http->ssl == 0);
1959 * Downgrade http version from 1.1 to 1.0
1960 * if +downgrade action applies.
1962 if ( (csp->action->flags & ACTION_DOWNGRADE)
1963 && (!strcmpic(http->ver, "HTTP/1.1")))
1966 http->ver = strdup("HTTP/1.0");
1968 if (http->ver == NULL)
1970 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1975 * Rebuild the request line.
1977 freez(*request_line);
1978 *request_line = strdup(http->gpc);
1979 string_append(request_line, " ");
1981 if (fwd->forward_host)
1983 string_append(request_line, http->url);
1987 string_append(request_line, http->path);
1989 string_append(request_line, " ");
1990 string_append(request_line, http->ver);
1992 if (*request_line == NULL)
1994 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1996 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
2000 /*********************************************************************
2002 * Function : change_request_destination
2004 * Description : Parse a (rewritten) request line and regenerate
2005 * the http request data.
2008 * 1 : csp = Current client state (buffers, headers, etc...)
2010 * Returns : Forwards the parse_http_request() return code.
2011 * Terminates in case of memory problems.
2013 *********************************************************************/
2014 static jb_err change_request_destination(struct client_state *csp)
2016 struct http_request *http = csp->http;
2019 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
2020 free_http_request(http);
2021 err = parse_http_request(csp->headers->first->str, http, csp);
2022 if (JB_ERR_OK != err)
2024 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
2025 jb_err_to_string(err));
2029 /* XXX: ocmd is a misleading name */
2030 http->ocmd = strdup(http->cmd);
2031 if (http->ocmd == NULL)
2033 log_error(LOG_LEVEL_FATAL,
2034 "Out of memory copying rewritten HTTP request line");
2042 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2043 /*********************************************************************
2045 * Function : server_response_is_complete
2047 * Description : Determines whether we should stop reading
2048 * from the server socket.
2051 * 1 : csp = Current client state (buffers, headers, etc...)
2052 * 2 : content_length = Length of content received so far.
2054 * Returns : TRUE if the response is complete,
2057 *********************************************************************/
2058 static int server_response_is_complete(struct client_state *csp, size_t content_length)
2060 int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2062 if (!strcmpic(csp->http->gpc, "HEAD"))
2065 * "HEAD" implies no body, we are thus expecting
2066 * no content. XXX: incomplete "list" of methods?
2068 csp->expected_content_length = 0;
2069 content_length_known = TRUE;
2072 if (csp->http->status == 304)
2075 * Expect no body. XXX: incomplete "list" of status codes?
2077 csp->expected_content_length = 0;
2078 content_length_known = TRUE;
2081 return (content_length_known && ((0 == csp->expected_content_length)
2082 || (csp->expected_content_length <= content_length)));
2084 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2086 /*********************************************************************
2088 * Function : mark_server_socket_tainted
2090 * Description : Makes sure we don't reuse a server socket
2091 * (if we didn't read everything the server sent
2092 * us reusing the socket would lead to garbage).
2095 * 1 : csp = Current client state (buffers, headers, etc...)
2099 *********************************************************************/
2100 static void mark_server_socket_tainted(struct client_state *csp)
2102 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2104 log_error(LOG_LEVEL_CONNECT, "Unsetting keep-alive flag.");
2105 csp->flags &= ~CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE;
2109 /*********************************************************************
2111 * Function : get_request_line
2113 * Description : Read the client request line.
2116 * 1 : csp = Current client state (buffers, headers, etc...)
2118 * Returns : Pointer to request line or NULL in case of errors.
2120 *********************************************************************/
2121 static char *get_request_line(struct client_state *csp)
2123 char buf[BUFFER_SIZE];
2124 char *request_line = NULL;
2127 memset(buf, 0, sizeof(buf));
2131 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2133 if (len <= 0) return NULL;
2136 * If there is no memory left for buffering the
2137 * request, there is nothing we can do but hang up
2139 if (add_to_iob(csp, buf, len))
2144 request_line = get_header(csp->iob);
2146 } while ((NULL != request_line) && ('\0' == *request_line));
2148 return request_line;
2153 /*********************************************************************
2155 * Function : receive_client_request
2157 * Description : Read the client's request (more precisely the
2158 * client headers) and answer it if necessary.
2160 * Note that since we're not using select() we could get
2161 * blocked here if a client connected, then didn't say
2165 * 1 : csp = Current client state (buffers, headers, etc...)
2167 * Returns : JB_ERR_OK, JB_ERR_PARSE or JB_ERR_MEMORY
2169 *********************************************************************/
2170 static jb_err receive_client_request(struct client_state *csp)
2172 char buf[BUFFER_SIZE];
2175 struct http_request *http;
2179 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2180 struct list header_list;
2181 struct list *headers = &header_list;
2185 memset(buf, 0, sizeof(buf));
2187 req = get_request_line(csp);
2189 if ((NULL != req) && ('\0' != *req))
2191 /* Request received. Validate and parse it. */
2193 /* Does the request line look invalid? */
2194 if (client_protocol_is_unsupported(csp, req))
2197 * Yes. The request has already been
2198 * answered with a error response, the buffers
2199 * were freed and we're done with chatting.
2201 return JB_ERR_PARSE;
2204 #ifdef FEATURE_FORCE_LOAD
2206 * If this request contains the FORCE_PREFIX and blocks
2207 * aren't enforced, get rid of it and set the force flag.
2209 if (strstr(req, FORCE_PREFIX))
2211 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2213 log_error(LOG_LEVEL_FORCE,
2214 "Ignored force prefix in request: \"%s\".", req);
2218 strclean(req, FORCE_PREFIX);
2219 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2220 csp->flags |= CSP_FLAG_FORCED;
2223 #endif /* def FEATURE_FORCE_LOAD */
2225 err = parse_http_request(req, http, csp);
2226 if (JB_ERR_OK != err)
2228 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2234 if (http->cmd == NULL)
2236 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2237 /* XXX: Use correct size */
2238 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2239 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2241 free_http_request(http);
2242 return JB_ERR_PARSE;
2245 /* grab the rest of the client's headers */
2249 p = get_header(csp->iob);
2253 /* There are no additional headers to read. */
2260 * We didn't receive a complete header
2261 * line yet, get the rest of it.
2263 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2266 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2267 destroy_list(headers);
2268 return JB_ERR_PARSE;
2271 if (add_to_iob(csp, buf, len))
2274 * If there is no memory left for buffering the
2275 * request, there is nothing we can do but hang up
2277 destroy_list(headers);
2278 return JB_ERR_MEMORY;
2284 * We were able to read a complete
2285 * header and can finaly enlist it.
2292 if (http->host == NULL)
2295 * If we still don't know the request destination,
2296 * the request is invalid or the client uses
2297 * Privoxy without its knowledge.
2299 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2302 * Our attempts to get the request destination
2303 * elsewhere failed or Privoxy is configured
2304 * to only accept proxy requests.
2306 * An error response has already been send
2307 * and we're done here.
2309 return JB_ERR_PARSE;
2314 * Determine the actions for this URL
2316 #ifdef FEATURE_TOGGLE
2317 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2319 /* Most compatible set of actions (i.e. none) */
2320 init_current_action(csp->action);
2323 #endif /* ndef FEATURE_TOGGLE */
2325 get_url_actions(csp, http);
2329 * Save a copy of the original request for logging
2331 http->ocmd = strdup(http->cmd);
2332 if (http->ocmd == NULL)
2334 log_error(LOG_LEVEL_FATAL,
2335 "Out of memory copying HTTP request line");
2337 enlist(csp->headers, http->cmd);
2339 /* Append the previously read headers */
2340 list_append_list_unique(csp->headers, headers);
2341 destroy_list(headers);
2348 /*********************************************************************
2350 * Function : parse_client_request
2352 * Description : Parses the client's request and decides what to do
2355 * Note that since we're not using select() we could get
2356 * blocked here if a client connected, then didn't say
2360 * 1 : csp = Current client state (buffers, headers, etc...)
2362 * Returns : JB_ERR_OK or JB_ERR_PARSE
2364 *********************************************************************/
2365 static jb_err parse_client_request(struct client_state *csp)
2367 struct http_request *http = csp->http;
2370 err = sed(csp, FILTER_CLIENT_HEADERS);
2371 if (JB_ERR_OK != err)
2373 /* XXX: Should be handled in sed(). */
2374 assert(err == JB_ERR_PARSE);
2375 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers.");
2377 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2379 /* Check request line for rewrites. */
2380 if ((NULL == csp->headers->first->str)
2381 || (strcmp(http->cmd, csp->headers->first->str) &&
2382 (JB_ERR_OK != change_request_destination(csp))))
2385 * A header filter broke the request line - bail out.
2387 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2388 /* XXX: Use correct size */
2389 log_error(LOG_LEVEL_CLF,
2390 "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2391 log_error(LOG_LEVEL_ERROR,
2392 "Invalid request line after applying header filters.");
2393 free_http_request(http);
2395 return JB_ERR_PARSE;
2403 /*********************************************************************
2407 * Description : Once a connection to the client has been accepted,
2408 * this function is called (via serve()) to handle the
2409 * main business of the communication. When this
2410 * function returns, the caller must close the client
2413 * FIXME: chat is nearly thousand lines long.
2417 * 1 : csp = Current client state (buffers, headers, etc...)
2419 * Returns : Nothing.
2421 *********************************************************************/
2422 static void chat(struct client_state *csp)
2424 char buf[BUFFER_SIZE];
2431 int ms_iis5_hack = 0;
2432 size_t byte_count = 0;
2433 int forwarded_connect_retries = 0;
2434 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2435 const struct forward_spec *fwd;
2436 struct http_request *http;
2437 int len; /* for buffer sizes (and negative error codes) */
2439 /* Function that does the content filtering for the current request */
2440 filter_function_ptr content_filter = NULL;
2442 /* Skeleton for HTTP response, if we should intercept the request */
2443 struct http_response *rsp;
2445 memset(buf, 0, sizeof(buf));
2449 if (receive_client_request(csp) != JB_ERR_OK)
2453 if (parse_client_request(csp) != JB_ERR_OK)
2458 /* decide how to route the HTTP request */
2459 fwd = forward_url(csp, http);
2462 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2463 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2467 * build the http request to send to the server
2468 * we have to do one of the following:
2470 * create = use the original HTTP request to create a new
2471 * HTTP request that has either the path component
2472 * without the http://domainspec (w/path) or the
2473 * full orininal URL (w/url)
2474 * Note that the path and/or the HTTP version may
2475 * have been altered by now.
2477 * connect = Open a socket to the host:port of the server
2478 * and short-circuit server and client socket.
2480 * pass = Pass the request unchanged if forwarding a CONNECT
2481 * request to a parent proxy. Note that we'll be sending
2482 * the CFAIL message ourselves if connecting to the parent
2483 * fails, but we won't send a CSUCCEED message if it works,
2484 * since that would result in a double message (ours and the
2485 * parent's). After sending the request to the parent, we simply
2488 * here's the matrix:
2491 * +--------+--------+
2493 * 0 | create | connect|
2495 * Forwarding +--------+--------+
2497 * 1 | create | pass |
2499 * +--------+--------+
2503 if (http->ssl && connect_port_is_forbidden(csp))
2505 const char *acceptable_connect_ports =
2506 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2507 assert(NULL != acceptable_connect_ports);
2508 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2509 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2510 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2511 csp->action->flags |= ACTION_BLOCK;
2517 freez(csp->headers->first->str);
2518 build_request_line(csp, fwd, &csp->headers->first->str);
2521 hdr = list_to_text(csp->headers);
2524 /* FIXME Should handle error properly */
2525 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2529 * We have a request. Check if one of the crunchers wants it.
2531 if (crunch_response_triggered(csp, crunchers_all))
2534 * Yes. The client got the crunch response
2535 * and we are done here after cleaning up.
2538 list_remove_all(csp->headers);
2544 * The headers can't be removed earlier because
2545 * they were still needed for the referrer check
2546 * in case of CGI crunches.
2548 * XXX: Would it be worth to move the referrer check
2549 * into client_referrer() and set a flag if it's trusted?
2551 list_remove_all(csp->headers);
2553 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2555 if (fwd->forward_host)
2557 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2558 fwd->forward_host, fwd->forward_port, http->hostport);
2562 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2565 /* here we connect to the server, gateway, or the forwarder */
2567 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2568 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2570 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2571 forwarded_connect_retries, http->hostport);
2574 if (csp->sfd == JB_INVALID_SOCKET)
2576 if (fwd->type != SOCKS_NONE)
2579 rsp = error_response(csp, "forwarding-failed", errno);
2581 else if (errno == EINVAL)
2583 rsp = error_response(csp, "no-such-domain", errno);
2587 rsp = error_response(csp, "connect-failed", errno);
2588 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2592 /* Write the answer to the client */
2595 send_crunch_response(csp, rsp);
2602 if (fwd->forward_host || (http->ssl == 0))
2605 * Write the client's (modified) header to the server
2606 * (along with anything else that may be in the buffer)
2608 if (write_socket(csp->sfd, hdr, strlen(hdr))
2609 || (flush_socket(csp->sfd, csp->iob) < 0))
2611 log_error(LOG_LEVEL_CONNECT,
2612 "write header to: %s failed: %E", http->hostport);
2614 rsp = error_response(csp, "connect-failed", errno);
2617 send_crunch_response(csp, rsp);
2627 * We're running an SSL tunnel and we're not forwarding,
2628 * so just send the "connect succeeded" message to the
2629 * client, flush the rest, and get out of the way.
2631 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2639 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2641 /* we're finished with the client's header */
2644 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2646 /* pass data between the client and server
2647 * until one or the other shuts down the connection.
2656 * FD_ZERO here seems to point to an errant macro which crashes.
2657 * So do this by hand for now...
2659 memset(&rfds,0x00,sizeof(fd_set));
2663 FD_SET(csp->cfd, &rfds);
2664 FD_SET(csp->sfd, &rfds);
2666 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2667 if (server_body && server_response_is_complete(csp, byte_count))
2669 log_error(LOG_LEVEL_CONNECT,
2670 "Done reading from server. Expected content length: %d. "
2671 "Actual content length: %d. Most recently received: %d.",
2672 csp->expected_content_length, byte_count, len);
2675 * XXX: should not jump around,
2676 * chat() is complicated enough already.
2680 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2682 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2686 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2687 mark_server_socket_tainted(csp);
2692 * This is the body of the browser's request,
2693 * just read and write it.
2695 if (FD_ISSET(csp->cfd, &rfds))
2697 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2701 /* XXX: not sure if this is necessary. */
2702 mark_server_socket_tainted(csp);
2703 break; /* "game over, man" */
2706 if (write_socket(csp->sfd, buf, (size_t)len))
2708 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2709 mark_server_socket_tainted(csp);
2716 * The server wants to talk. It could be the header or the body.
2717 * If `hdr' is null, then it's the header otherwise it's the body.
2718 * FIXME: Does `hdr' really mean `host'? No.
2720 if (FD_ISSET(csp->sfd, &rfds))
2723 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2727 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2729 if (http->ssl && (fwd->forward_host == NULL))
2732 * Just hang up. We already confirmed the client's CONNECT
2733 * request with status code 200 and unencrypted content is
2734 * no longer welcome.
2736 log_error(LOG_LEVEL_ERROR,
2737 "CONNECT already confirmed. Unable to tell the client about the problem.");
2740 else if (byte_count)
2743 * Just hang up. We already transmitted the original headers
2744 * and parts of the original content and therefore missed the
2745 * chance to send an error message (without risking data corruption).
2747 * XXX: we could retry with a fancy range request here.
2749 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2750 "Unable to tell the client about the problem.");
2751 mark_server_socket_tainted(csp);
2755 rsp = error_response(csp, "connect-failed", errno);
2758 send_crunch_response(csp, rsp);
2764 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2765 if (csp->flags & CSP_FLAG_CHUNKED)
2767 if ((len >= 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
2769 /* XXX: this is a temporary hack */
2770 log_error(LOG_LEVEL_CONNECT,
2771 "Looks like we reached the end of the last chunk. "
2772 "We better stop reading.");
2773 csp->expected_content_length = byte_count + (size_t)len;
2774 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2778 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2781 * Add a trailing zero to let be able to use string operations.
2782 * XXX: do we still need this with filter_popups gone?
2787 * Normally, this would indicate that we've read
2788 * as much as the server has sent us and we can
2789 * close the client connection. However, Microsoft
2790 * in its wisdom has released IIS/5 with a bug that
2791 * prevents it from sending the trailing \r\n in
2792 * a 302 redirect header (and possibly other headers).
2793 * To work around this if we've haven't parsed
2794 * a full header we'll append a trailing \r\n
2795 * and see if this now generates a valid one.
2797 * This hack shouldn't have any impacts. If we've
2798 * already transmitted the header or if this is a
2799 * SSL connection, then we won't bother with this
2800 * hack. So we only work on partially received
2801 * headers. If we append a \r\n and this still
2802 * doesn't generate a valid header, then we won't
2803 * transmit anything to the client.
2808 if (server_body || http->ssl)
2811 * If we have been buffering up the document,
2812 * now is the time to apply content modification
2813 * and send the result to the client.
2817 p = execute_content_filter(csp, content_filter);
2819 * If the content filter fails, use the original
2820 * buffer and length.
2821 * (see p != NULL ? p : csp->iob->cur below)
2825 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2828 if (JB_ERR_OK != update_server_headers(csp))
2830 log_error(LOG_LEVEL_FATAL,
2831 "Failed to update server headers. after filtering.");
2834 hdr = list_to_text(csp->headers);
2837 /* FIXME Should handle error properly */
2838 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2841 if (write_socket(csp->cfd, hdr, strlen(hdr))
2842 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2844 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2847 mark_server_socket_tainted(csp);
2855 break; /* "game over, man" */
2859 * This is NOT the body, so
2860 * Let's pretend the server just sent us a blank line.
2862 snprintf(buf, sizeof(buf), "\r\n");
2863 len = (int)strlen(buf);
2866 * Now, let the normal header parsing algorithm below do its
2867 * job. If it fails, we'll exit instead of continuing.
2874 * If this is an SSL connection or we're in the body
2875 * of the server document, just write it to the client,
2876 * unless we need to buffer the body for later content-filtering
2878 if (server_body || http->ssl)
2883 * If there is no memory left for buffering the content, or the buffer limit
2884 * has been reached, switch to non-filtering mode, i.e. make & write the
2885 * header, flush the iob and buf, and get out of the way.
2887 if (add_to_iob(csp, buf, len))
2892 log_error(LOG_LEVEL_INFO,
2893 "Flushing header and buffers. Stepping back from filtering.");
2895 hdr = list_to_text(csp->headers);
2899 * Memory is too tight to even generate the header.
2900 * Send our static "Out-of-memory" page.
2902 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2903 rsp = cgi_error_memory();
2904 send_crunch_response(csp, rsp);
2905 mark_server_socket_tainted(csp);
2908 hdrlen = strlen(hdr);
2910 if (write_socket(csp->cfd, hdr, hdrlen)
2911 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2912 || (write_socket(csp->cfd, buf, (size_t)len)))
2914 log_error(LOG_LEVEL_CONNECT,
2915 "Flush header and buffers to client failed: %E");
2917 mark_server_socket_tainted(csp);
2922 * Reset the byte_count to the amount of bytes
2923 * we just flushed. len will be added a few lines below,
2924 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2926 byte_count = (size_t)flushed;
2928 content_filter = NULL;
2934 if (write_socket(csp->cfd, buf, (size_t)len))
2936 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2937 mark_server_socket_tainted(csp);
2941 byte_count += (size_t)len;
2946 const char *header_start;
2948 * We're still looking for the end of the server's header.
2949 * Buffer up the data we just read. If that fails, there's
2950 * little we can do but send our static out-of-memory page.
2952 if (add_to_iob(csp, buf, len))
2954 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2955 rsp = cgi_error_memory();
2956 send_crunch_response(csp, rsp);
2957 mark_server_socket_tainted(csp);
2961 header_start = csp->iob->cur;
2963 /* Convert iob into something sed() can digest */
2964 if (JB_ERR_PARSE == get_server_headers(csp))
2969 * Well, we tried our MS IIS/5 hack and it didn't work.
2970 * The header is incomplete and there isn't anything
2971 * we can do about it.
2973 log_error(LOG_LEVEL_INFO,
2974 "MS IIS5 hack didn't produce valid headers.");
2980 * Since we have to wait for more from the server before
2981 * we can parse the headers we just continue here.
2987 /* Did we actually get anything? */
2988 if (NULL == csp->headers->first)
2990 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2991 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2992 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2993 free_http_request(http);
2994 mark_server_socket_tainted(csp);
2998 assert(csp->headers->first->str);
3000 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
3001 strncmpic(csp->headers->first->str, "ICY", 3))
3004 * It doesn't look like a HTTP (or Shoutcast) response:
3005 * tell the client and log the problem.
3007 if (strlen(csp->headers->first->str) > 30)
3009 csp->headers->first->str[30] = '\0';
3011 log_error(LOG_LEVEL_ERROR,
3012 "Invalid server or forwarder response. Starts with: %s",
3013 csp->headers->first->str);
3014 log_error(LOG_LEVEL_CLF,
3015 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3016 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3017 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3018 free_http_request(http);
3019 mark_server_socket_tainted(csp);
3024 * We have now received the entire server header,
3025 * filter it and send the result to the client
3027 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
3029 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
3031 hdr = list_to_text(csp->headers);
3034 /* FIXME Should handle error properly */
3035 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3038 if (crunch_response_triggered(csp, crunchers_light))
3041 * One of the tags created by a server-header
3042 * tagger triggered a crunch. We already
3043 * delivered the crunch response to the client
3044 * and are done here after cleaning up.
3047 mark_server_socket_tainted(csp);
3050 /* Buffer and pcrs filter this if appropriate. */
3052 if (!http->ssl) /* We talk plaintext */
3054 content_filter = get_filter_function(csp);
3057 * Only write if we're not buffering for content modification
3059 if (!content_filter)
3062 * Write the server's (modified) header to
3063 * the client (along with anything else that
3064 * may be in the buffer)
3067 if (write_socket(csp->cfd, hdr, strlen(hdr))
3068 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
3070 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
3073 * The write failed, so don't bother mentioning it
3074 * to the client... it probably can't hear us anyway.
3077 mark_server_socket_tainted(csp);
3081 byte_count += (size_t)len;
3086 * XXX: the header lenght should probably
3087 * be calculated by get_server_headers().
3089 int header_length = csp->iob->cur - header_start;
3090 assert(csp->iob->cur > header_start);
3091 byte_count += (size_t)(len - header_length);
3094 /* we're finished with the server's header */
3100 * If this was a MS IIS/5 hack then it means the server
3101 * has already closed the connection. Nothing more to read.
3106 log_error(LOG_LEVEL_INFO,
3107 "Closed server connection detected with MS IIS5 hack enabled.");
3113 mark_server_socket_tainted(csp);
3114 return; /* huh? we should never get here */
3117 if (csp->content_length == 0)
3120 * If Privoxy didn't recalculate the Content-Lenght,
3121 * byte_count is still correct.
3123 csp->content_length = byte_count;
3126 if ((csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
3127 && (csp->expected_content_length != byte_count))
3129 log_error(LOG_LEVEL_ERROR,
3130 "Received %d bytes while expecting %d.",
3131 byte_count, csp->expected_content_length);
3132 mark_server_socket_tainted(csp);
3135 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
3136 csp->ip_addr_str, http->ocmd, csp->content_length);
3140 /*********************************************************************
3144 * Description : This is little more than chat. We only "serve" to
3145 * to close any socket that chat may have opened.
3148 * 1 : csp = Current client state (buffers, headers, etc...)
3152 *********************************************************************/
3154 void serve(struct client_state *csp)
3155 #else /* ifndef AMIGA */
3156 static void serve(struct client_state *csp)
3157 #endif /* def AMIGA */
3160 close_socket(csp->cfd);
3162 if (csp->sfd != JB_INVALID_SOCKET)
3164 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3165 if ((csp->config->feature_flags & RUNTIME_FEATURE_CONNECTION_KEEP_ALIVE)
3166 && (csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
3168 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
3172 forget_connection(csp->sfd);
3173 close_socket(csp->sfd);
3176 close_socket(csp->sfd);
3177 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3180 csp->flags &= ~CSP_FLAG_ACTIVE;
3186 /*********************************************************************
3188 * Function : server_thread
3190 * Description : We only exist to call `serve' in a threaded environment.
3193 * 1 : data = Current client state (buffers, headers, etc...)
3195 * Returns : Always 0.
3197 *********************************************************************/
3198 static int32 server_thread(void *data)
3200 serve((struct client_state *) data);
3207 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3208 /*********************************************************************
3212 * Description : Print usage info & exit.
3214 * Parameters : Pointer to argv[0] for identifying ourselves
3218 *********************************************************************/
3219 static void usage(const char *myname)
3221 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3225 #endif /* defined(unix) */
3228 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3229 #endif /* defined(unix) */
3230 "[--version] [configfile]\n"
3231 "Aborting\n", myname);
3236 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3239 #ifdef MUTEX_LOCKS_AVAILABLE
3240 /*********************************************************************
3242 * Function : privoxy_mutex_lock
3244 * Description : Locks a mutex.
3247 * 1 : mutex = The mutex to lock.
3249 * Returns : Void. May exit in case of errors.
3251 *********************************************************************/
3252 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3254 #ifdef FEATURE_PTHREAD
3255 int err = pthread_mutex_lock(mutex);
3258 if (mutex != &log_mutex)
3260 log_error(LOG_LEVEL_FATAL,
3261 "Mutex locking failed: %s.\n", strerror(err));
3266 EnterCriticalSection(mutex);
3267 #endif /* def FEATURE_PTHREAD */
3271 /*********************************************************************
3273 * Function : privoxy_mutex_unlock
3275 * Description : Unlocks a mutex.
3278 * 1 : mutex = The mutex to unlock.
3280 * Returns : Void. May exit in case of errors.
3282 *********************************************************************/
3283 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3285 #ifdef FEATURE_PTHREAD
3286 int err = pthread_mutex_unlock(mutex);
3289 if (mutex != &log_mutex)
3291 log_error(LOG_LEVEL_FATAL,
3292 "Mutex unlocking failed: %s.\n", strerror(err));
3297 LeaveCriticalSection(mutex);
3298 #endif /* def FEATURE_PTHREAD */
3302 /*********************************************************************
3304 * Function : privoxy_mutex_init
3306 * Description : Prepares a mutex.
3309 * 1 : mutex = The mutex to initialize.
3311 * Returns : Void. May exit in case of errors.
3313 *********************************************************************/
3314 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3316 #ifdef FEATURE_PTHREAD
3317 int err = pthread_mutex_init(mutex, 0);
3320 printf("Fatal error. Mutex initialization failed: %s.\n",
3325 InitializeCriticalSection(mutex);
3326 #endif /* def FEATURE_PTHREAD */
3328 #endif /* def MUTEX_LOCKS_AVAILABLE */
3330 /*********************************************************************
3332 * Function : initialize_mutexes
3334 * Description : Prepares mutexes if mutex support is available.
3338 * Returns : Void, exits in case of errors.
3340 *********************************************************************/
3341 static void initialize_mutexes(void)
3343 #ifdef MUTEX_LOCKS_AVAILABLE
3345 * Prepare global mutex semaphores
3347 privoxy_mutex_init(&log_mutex);
3348 privoxy_mutex_init(&log_init_mutex);
3349 privoxy_mutex_init(&connection_reuse_mutex);
3352 * XXX: The assumptions below are a bit naive
3353 * and can cause locks that aren't necessary.
3355 * For example older FreeBSD versions (< 6.x?)
3356 * have no gethostbyname_r, but gethostbyname is
3359 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3360 privoxy_mutex_init(&resolver_mutex);
3361 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3363 * XXX: should we use a single mutex for
3364 * localtime() and gmtime() as well?
3366 #ifndef HAVE_GMTIME_R
3367 privoxy_mutex_init(&gmtime_mutex);
3368 #endif /* ndef HAVE_GMTIME_R */
3370 #ifndef HAVE_LOCALTIME_R
3371 privoxy_mutex_init(&localtime_mutex);
3372 #endif /* ndef HAVE_GMTIME_R */
3375 privoxy_mutex_init(&rand_mutex);
3376 #endif /* ndef HAVE_RANDOM */
3377 #endif /* def MUTEX_LOCKS_AVAILABLE */
3381 /*********************************************************************
3385 * Description : Load the config file and start the listen loop.
3386 * This function is a lot more *sane* with the `load_config'
3387 * and `listen_loop' functions; although it stills does
3388 * a *little* too much for my taste.
3391 * 1 : argc = Number of parameters (including $0).
3392 * 2 : argv = Array of (char *)'s to the parameters.
3394 * Returns : 1 if : can't open config file, unrecognized directive,
3395 * stats requested in multi-thread mode, can't open the
3396 * log file, can't open the jar file, listen port is invalid,
3397 * any load fails, and can't bind port.
3399 * Else main never returns, the process must be signaled
3400 * to terminate execution. Or, on Windows, use the
3401 * "File", "Exit" menu option.
3403 *********************************************************************/
3405 int real_main(int argc, const char *argv[])
3407 int main(int argc, const char *argv[])
3411 unsigned int random_seed;
3413 struct passwd *pw = NULL;
3414 struct group *grp = NULL;
3417 char *pre_chroot_nslookup_to_load_resolver = NULL;
3424 #if !defined(_WIN32)
3432 * Parse the command line arguments
3434 * XXX: simply printing usage information in case of
3435 * invalid arguments isn't particularly user friendly.
3437 while (++argc_pos < argc)
3440 /* Check to see if the service must be installed or uninstalled */
3441 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3443 const char *pName = argv[argc_pos] + 9;
3446 exit( (install_service(pName)) ? 0 : 1 );
3448 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3450 const char *pName = argv[argc_pos] + 11;
3453 exit((uninstall_service(pName)) ? 0 : 1);
3455 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3457 bRunAsService = TRUE;
3458 w32_set_service_cwd();
3459 atexit(w32_service_exit_notify);
3462 #endif /* defined(_WIN32) */
3465 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3467 if (strcmp(argv[argc_pos], "--help") == 0)
3472 else if(strcmp(argv[argc_pos], "--version") == 0)
3474 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3480 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3485 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3487 if (++argc_pos == argc) usage(argv[0]);
3488 pidfile = strdup(argv[argc_pos]);
3491 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3493 if (++argc_pos == argc) usage(argv[argc_pos]);
3495 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3498 if (NULL == (grp = getgrnam(p)))
3500 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3504 if (NULL == (pw = getpwnam(argv[argc_pos])))
3506 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3509 if (p != NULL) *--p = '\0';
3512 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3514 if (++argc_pos == argc) usage(argv[0]);
3515 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3518 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3522 #endif /* defined(unix) */
3524 else if (argc_pos + 1 != argc)
3527 * This is neither the last command line
3528 * option, nor was it recognized before,
3529 * therefore it must be invalid.
3535 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3537 configfile = argv[argc_pos];
3540 } /* -END- while (more arguments) */
3543 if ( *configfile != '/' )
3545 char cwd[BUFFER_SIZE];
3547 size_t abs_file_size;
3549 /* make config-filename absolute here */
3550 if (NULL == getcwd(cwd, sizeof(cwd)))
3552 perror("failed to get current working directory");
3557 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3558 basedir = strdup(cwd);
3560 if (NULL == basedir ||
3561 NULL == (abs_file = malloc(abs_file_size)))
3563 perror("malloc failed");
3566 strlcpy(abs_file, basedir, abs_file_size);
3567 strlcat(abs_file, "/", abs_file_size );
3568 strlcat(abs_file, configfile, abs_file_size);
3569 configfile = abs_file;
3571 #endif /* defined unix */
3575 clients->next = NULL;
3577 /* XXX: factor out initialising after the next stable release. */
3580 #elif defined(_WIN32)
3584 /* Prepare mutexes if supported and necessary. */
3585 initialize_mutexes();
3587 /* Enable logging until further notice. */
3588 init_log_module(Argv[0]);
3590 random_seed = (unsigned int)time(NULL);
3592 srandom(random_seed);
3595 #endif /* ifdef HAVE_RANDOM */
3598 * Unix signal handling
3600 * Catch the abort, interrupt and terminate signals for a graceful exit
3601 * Catch the hangup signal so the errlog can be reopened.
3602 * Ignore the broken pipe signals (FIXME: Why?)
3604 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3607 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3608 const int ignored_signals[] = { SIGPIPE, 0 };
3610 for (idx = 0; catched_signals[idx] != 0; idx++)
3612 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3613 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3615 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3616 #endif /* ifdef sun */
3618 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3622 for (idx = 0; ignored_signals[idx] != 0; idx++)
3624 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3626 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3631 #else /* ifdef _WIN32 */
3632 # ifdef _WIN_CONSOLE
3634 * We *are* in a windows console app.
3635 * Print a verbose messages about FAQ's and such
3637 printf("%s", win32_blurb);
3638 # endif /* def _WIN_CONSOLE */
3639 #endif /* def _WIN32 */
3642 /* Initialize the CGI subsystem */
3643 cgi_init_error_messages();
3646 * If runnig on unix and without the --nodaemon
3647 * option, become a daemon. I.e. fork, detach
3648 * from tty and get process group leadership
3661 if ( pid < 0 ) /* error */
3666 else if ( pid != 0 ) /* parent */
3671 * must check for errors
3672 * child died due to missing files aso
3675 wpid = waitpid( pid, &status, WNOHANG );
3684 /* Should be more portable, but not as well tested */
3689 #else /* ndef __FreeBSD__ */
3691 #endif /* ndef __FreeBSD__ */
3692 fd = open("/dev/tty", O_RDONLY);
3695 /* no error check here */
3696 ioctl( fd, TIOCNOTTY,0 );
3701 * stderr (fd 2) will be closed later on, when the
3702 * log file has been parsed.
3709 } /* -END- if (!no_daemon) */
3712 * As soon as we have written the PID file, we can switch
3713 * to the user and group ID indicated by the --user option
3719 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3721 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3727 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3729 /* Read the time zone file from /etc before doing chroot. */
3731 if (NULL != pre_chroot_nslookup_to_load_resolver
3732 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3734 /* Initialize resolver library. */
3735 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3737 if (chroot(pw->pw_dir) < 0)
3739 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3743 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3746 if (setuid(pw->pw_uid))
3748 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3752 char putenv_dummy[64];
3754 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3755 if (putenv(putenv_dummy) != 0)
3757 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3760 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3761 if (putenv(putenv_dummy) != 0)
3763 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3769 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3772 #endif /* defined unix */
3775 /* This will be FALSE unless the command line specified --service
3779 /* Yup, so now we must attempt to establish a connection
3780 * with the service dispatcher. This will only work if this
3781 * process was launched by the service control manager to
3782 * actually run as a service. If this isn't the case, i've
3783 * known it take around 30 seconds or so for the call to return.
3786 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3787 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3789 /* Service has run, and at this point is now being stopped, so just return */