1 const char jcc_rcs[] = "$Id: jcc.c,v 1.243 2009/04/17 11:27:49 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) 2001-2009 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 1.243 2009/04/17 11:27:49 fabiankeil
37 * Petr Pisar's privoxy-3.0.12-ipv6-3.diff.
39 * Revision 1.242 2009/04/11 10:44:47 fabiankeil
40 * Update a comment. We're not in Kansas anymore.
42 * Revision 1.241 2009/04/11 10:37:23 fabiankeil
43 * When dropping connections due to ACL, don't leak csp->ip_addr_str.
45 * Revision 1.240 2009/04/09 10:12:54 fabiankeil
46 * Fix two cases in which an invalid server response would result
47 * in the client connection being closed without sending an error
50 * Revision 1.239 2009/04/07 11:43:50 fabiankeil
51 * If the server rudely resets the connection directly after sending the
52 * headers, pass the mess to the client instead of sending an incorrect
53 * connect-failed message. Fixes #2698674 reported by mybugaccount.
55 * Revision 1.238 2009/03/27 14:42:30 fabiankeil
56 * Correct the status code for CONNECTION_TIMEOUT_RESPONSE.
58 * Revision 1.237 2009/03/27 14:32:04 fabiankeil
59 * If spawning a child in listen_loop() fails, send a real
60 * HTTP response to the client and continue listening for
61 * new connections without artificial delay.
63 * Revision 1.236 2009/03/25 17:30:24 fabiankeil
64 * In serve(), keep the client socket open until we marked the
65 * server socket as unused. This should increase the chances
66 * that we reuse the connection for the client's next request
67 * to the same destination.
69 * Revision 1.235 2009/03/18 21:01:20 fabiankeil
70 * Comment fix. Spotted by Roland.
72 * Revision 1.234 2009/03/18 20:48:42 fabiankeil
73 * If the --no-daemon option is used, enable LOG_LEVEL_INFO
74 * before the config file has been parsed (as we always did).
76 * Revision 1.233 2009/03/13 14:10:07 fabiankeil
77 * Fix some more harmless warnings on amd64.
79 * Revision 1.232 2009/03/08 19:29:16 fabiankeil
80 * Reinitialize the timeout structure every time before passing
81 * it to select(). Apparently some implementations mess with it.
82 * Probably fixes #2669131 reported by cyberpatrol.
84 * Revision 1.231 2009/03/08 14:19:23 fabiankeil
85 * Fix justified (but harmless) compiler warnings
86 * on platforms where sizeof(int) < sizeof(long).
88 * Revision 1.230 2009/03/07 13:09:17 fabiankeil
89 * Change csp->expected_content and_csp->expected_content_length from
90 * size_t to unsigned long long to reduce the likelihood of integer
91 * overflows that would let us close the connection prematurely.
92 * Bug found while investigating #2669131, reported by cyberpatrol.
94 * Revision 1.229 2009/03/07 11:17:01 fabiankeil
95 * Fix compiler warning.
97 * Revision 1.228 2009/03/06 20:30:13 fabiankeil
98 * Log unsigned values as such.
100 * Revision 1.227 2009/03/02 19:18:11 fabiankeil
101 * Streamline parse_http_request()'s prototype. As
102 * cparser pointed out it doesn't actually use csp.
104 * Revision 1.226 2009/03/01 18:28:24 fabiankeil
105 * Help clang understand that we aren't dereferencing
106 * NULL pointers here.
108 * Revision 1.225 2009/02/19 18:09:32 fabiankeil
109 * Unbreak build without FEATURE_CONNECTION_KEEP_ALIVE.
112 * Revision 1.224 2009/02/14 15:32:04 fabiankeil
113 * Add the request URL to the timeout message in chat().
116 * Revision 1.223 2009/02/09 21:21:16 fabiankeil
117 * Now that init_log_module() is called earlier, call show_version()
118 * later on from main() directly so it doesn't get called for --help
121 * Revision 1.222 2009/02/08 12:56:51 fabiankeil
122 * Call initialize_mutexes() before init_log_module() again.
123 * Broken since r220, might be the cause of Lee's #2579448.
125 * Revision 1.221 2009/02/06 18:02:58 fabiankeil
126 * When dropping privileges, also give up membership in supplementary
127 * groups. Thanks to Matthias Drochner for reporting the problem,
128 * providing the initial patch and testing the final version.
130 * Revision 1.220 2009/02/04 18:29:07 fabiankeil
131 * Initialize the log module before parsing arguments.
132 * Thanks to Matthias Drochner for the report.
134 * Revision 1.219 2009/01/31 16:08:21 fabiankeil
135 * Remove redundant error check in receive_client_request().
137 * Revision 1.218 2009/01/31 12:25:54 fabiankeil
138 * Flatten indentation in receive_client_request().
140 * Revision 1.217 2009/01/07 19:50:09 fabiankeil
141 * - If the socket-timeout has been reached and the client
142 * hasn't received any data yet, send an explanation before
143 * closing the connection.
144 * - In get_request_line(), signal timeouts the right way.
146 * Revision 1.216 2008/12/24 22:13:11 ler762
147 * fix GCC 3.4.4 warning
149 * Revision 1.215 2008/12/24 17:06:19 fabiankeil
150 * Keep a thread around to timeout alive connections
151 * even if no new requests are coming in.
153 * Revision 1.214 2008/12/20 14:53:55 fabiankeil
154 * Add config option socket-timeout to control the time
155 * Privoxy waits for data to arrive on a socket. Useful
156 * in case of stale ssh tunnels or when fuzz-testing.
158 * Revision 1.213 2008/12/15 18:45:51 fabiankeil
159 * When logging crunches, log the whole URL, so one can easily
160 * differentiate between vanilla HTTP and CONNECT requests.
162 * Revision 1.212 2008/12/14 15:46:22 fabiankeil
163 * Give crunched requests their own log level.
165 * Revision 1.211 2008/12/06 10:05:03 fabiankeil
166 * Downgrade "Received x bytes while expecting y." message to
167 * LOG_LEVEL_CONNECT as it doesn't necessarily indicate an error.
169 * Revision 1.210 2008/12/02 22:03:18 fabiankeil
170 * Don't miscalculate byte_count if we don't get all the
171 * server headers with one read_socket() call. With keep-alive
172 * support enabled, this caused delays until the server closed
175 * Revision 1.209 2008/11/27 09:44:04 fabiankeil
176 * Cosmetics for the last commit: Don't watch out for
177 * the last chunk if the content isn't chunk-encoded or
178 * if we already determined the content length previously.
180 * Revision 1.208 2008/11/26 18:24:17 fabiankeil
181 * Recognize that the server response is complete if the
182 * last chunk is read together with the server headers.
185 * Revision 1.207 2008/11/25 17:25:16 fabiankeil
186 * Don't convert the client-header list to text until we need to.
188 * Revision 1.206 2008/11/23 17:00:11 fabiankeil
189 * Some more chat() cosmetics.
191 * Revision 1.205 2008/11/16 12:43:49 fabiankeil
192 * Turn keep-alive support into a runtime feature
193 * that is disabled by setting keep-alive-timeout
194 * to a negative value.
196 * Revision 1.204 2008/11/06 19:42:17 fabiankeil
197 * Fix last-chunk detection hack to also apply
198 * if buf[] contains nothing but the last-chunk.
200 * Revision 1.203 2008/11/06 18:34:35 fabiankeil
201 * Factor receive_client_request() and
202 * parse_client_request() out of chat().
204 * Revision 1.202 2008/11/02 18:40:34 fabiankeil
205 * If we received a different amount of data than we expected,
206 * log a warning and make sure the server socket isn't reused.
208 * Revision 1.201 2008/11/02 16:48:20 fabiankeil
209 * Revert revision 1.195 and try again.
211 * Revision 1.200 2008/10/26 16:53:18 fabiankeil
214 * Revision 1.199 2008/10/26 15:36:10 fabiankeil
215 * Remove two debug messages with LOG_LEVEL_INFO.
217 * Revision 1.198 2008/10/22 15:19:55 fabiankeil
218 * Once More, With Feeling: if there is no logfile
219 * because the user didn't specify one, we shouldn't
220 * call init_error_log() after receiving SIGHUP either.
222 * Revision 1.197 2008/10/20 17:02:40 fabiankeil
223 * If SIGHUP is received while we aren't running in daemon
224 * mode, calling init_error_log() would be a mistake.
226 * Revision 1.196 2008/10/16 09:16:41 fabiankeil
227 * - Fix two gcc44 conversion warnings.
228 * - Don't bother logging the last five bytes
231 * Revision 1.195 2008/10/13 16:04:37 fabiankeil
232 * Make sure we don't try to reuse tainted server sockets.
234 * Revision 1.194 2008/10/12 18:35:18 fabiankeil
235 * The last commit was a bit too ambitious, apparently the content
236 * length adjustment is only necessary if we aren't buffering.
238 * Revision 1.193 2008/10/12 15:57:35 fabiankeil
239 * Fix content length calculation if we read headers
240 * and the start of the body at once. Now that we have
241 * FEATURE_CONNECTION_KEEP_ALIVE, it actually matters.
243 * Revision 1.192 2008/10/11 18:19:14 fabiankeil
244 * Even more chat() cosmetics.
246 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
247 * Reformat some comments in chat().
249 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
250 * In case of chunk-encoded content, stop reading if
251 * the buffer looks like it ends with the last chunk.
253 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
254 * Let server_response_is_complete() deal properly with
255 * content that is neither buffered nor read all at once.
257 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
258 * Flush work-in-progress changes to keep outgoing connections
259 * alive where possible. Incomplete and mostly #ifdef'd out.
261 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
262 * Add mutex lock support for _WIN32.
264 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
265 * Prepare for critical sections on Windows by adding a
266 * layer of indirection before the pthread mutex functions.
268 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
269 * Remove FEATURE_COOKIE_JAR.
271 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
272 * - Silence LLVM/Clang complaint.
273 * - Make received_hup_signal static.
274 * - Hide definitions for basedir, pidfile and received_hup_signal
275 * from __EMX__ as they only seem to be used in case of #ifdef unix.
277 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
278 * Accept Shoutcast responses again. Problem reported
279 * and fix suggested by Stefan in #2062860.
281 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
282 * Fix possible NULL-pointer dereference reported
283 * by din_a4 in #2003937. Pointy hat to me.
285 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
286 * Streamline sed()'s prototype and declare
287 * the header parse and add structures static.
289 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
290 * - Mark csp as immutable for send_crunch_response().
291 * - Fix comment spelling.
293 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
294 * Factor update_server_headers() out of sed(), ditch the
295 * first_run hack and make server_patterns_light static.
297 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
298 * Don't provide get_header() with the whole client state
299 * structure when it only needs access to csp->iob.
301 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
302 * Make the "read the rest of the headers" loop a bit more readable.
304 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
305 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
306 * - Remove useless comment.
308 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
309 * Fix comment grammar.
311 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
312 * Remove the pointless buffer in client_protocol_is_unsupported().
314 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
315 * Least-effort fix for bug #1821930 (reported by Lee):
316 * If the response doesn't look like HTTP,
317 * tell the client and log the problem.
319 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
320 * Don't pass the whole csp structure to flush_socket()
321 * when it only needs a file descriptor and a buffer.
323 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
324 * Remove kill-popups action.
326 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
327 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
329 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
330 * Remove the treat-forbidden-connects-like-blocks action. We now
331 * use the "blocked" page for forbidden CONNECT requests by default.
333 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
334 * Also use shiny new connect_port_is_forbidden() in jcc.c.
336 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
337 * Rename url_actions() to get_url_actions() and let it
338 * use the standard parameter ordering.
340 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
341 * Let forward_url() use the standard parameter ordering
342 * and mark its second parameter immutable.
344 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
345 * Remove the "Listening ... for local connections only" log message.
346 * Whether or not remote connections are able to reach Privoxy is up
347 * to the operating system.
349 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
350 * Prevent the log messages for CONNECT requests to unacceptable
351 * ports from printing the limit-connect argument as [null] if
352 * limit-connect hasn't been explicitly enabled.
354 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
355 * Make sure all console-mode apps get a usage() instance
357 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
358 * Reword NO_SERVER_DATA_RESPONSE to make it harder
359 * to misunderstand what the message is all about.
361 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
362 * Unbreak trustfile which previously didn't work without
363 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
365 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
366 * Plug memory leak. Spotted by Valgrind, triggered by
367 * Privoxy-Regression-Test feeding proxyfuzz.py.
369 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
370 * In the HTTP snipplets, refer to the client as client.
372 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
373 * Emit a log message when activating the MS IIS5 hack.
375 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
376 * Log the "weak randomization factor" warning only
377 * once for mingw32 and provide some more details.
379 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
380 * Initialize log module after initializing mutexes, future
381 * deadlocks in that code should now work cross-platform.
383 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
384 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
385 * Reported by Bert van Leeuwen in BR#1818808.
387 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
388 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
390 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
391 * When in daemon mode, close stderr after the configuration file has been
392 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
394 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
395 * - Fix a crash when parsing invalid requests whose first header
396 * is rejected by get_header(). Regression (re?)introduced
397 * in r1.143 by yours truly.
398 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
399 * client_cookie_adder() to make sure send-vanilla-wafer can be
400 * controlled through tags (and thus regression-tested).
402 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
403 * - Move get_filter_function() from jcc.c to filters.c
404 * so the filter functions can be static.
405 * - Don't bother filtering body-less responses.
407 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
408 * Execute content filters through execute_content_filter().
410 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
411 * Initialize req to NULL to make sure it's defined if the
412 * first read_socket() call fails. Reported by icmp30.
414 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
415 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
416 * extensive comments moved to user manual.
418 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
419 * Don't crash if a broken header filter wiped out the request line.
421 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
422 * Fix byte_count calculation in case of flushes
423 * and don't parse the server headers a second time.
425 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
426 * - If there's a connection problem after we already forwarded
427 * parts of the original content, just hang up. Fixes BR#1776724.
428 * - Fix warnings about unused code on mingw32.
429 * - In case of flushes, calculate the byte count
430 * less incorrectly (I think).
432 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
433 * Add some more prototypes for static functions.
435 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
436 * Comment out request_contains_null_bytes() until it's used again.
438 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
439 * #1763173 from Stefan Huehner: s@const static@static const@
440 * and declare some more functions static.
442 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
443 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
444 * and explain why they were rejected in the first place.
445 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
446 * CONNECT requests. The request line was missing.
447 * - Add two more XXX reminders as we don't have enough already.
449 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
450 * As Hal noticed, checking dispatch_cgi() as the last cruncher
451 * looks like a bug if CGI requests are blocked unintentionally,
452 * so don't do it unless the user enabled the new config option
453 * "allow-cgi-request-crunching".
455 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
456 * - Allow to rewrite the request destination behind the client's back.
457 * - Turn the weird-looking unconditional for loop that
458 * reads the client request into a conditional while loop.
459 * Move the stuff that only runs once out of the loop.
460 * - Move parts of chat(), server_content_type() and the
461 * necessary stuff to fix BR#1750917 into get_filter_function().
463 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
464 * Temporary workaround for BR#1730105.
466 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
467 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
468 * deadlocks and crashes on OpenBSD and possibly other OS with neither
469 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
470 * Thanks to Ralf Horstmann for report and solution.
472 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
473 * Add forward-override{} to change the forwarding settings through
474 * action sections. This is mainly interesting to forward different
475 * clients differently (for example based on User-Agent or request
478 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
479 * - Let usage() mention the --chroot parameter.
480 * - Use read_socket() consistently and always leave
481 * the last buffer byte alone, even in cases where
482 * null termination (currently) doesn't matter.
484 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
485 * - Fix config file loading on Unix if no config file is specified.
486 * Since r1.97 Privoxy would always interpret the last argument as
487 * config file, even if it's a valid command line option.
488 * - Abort in case of unrecognized command line options. Closes #1719696.
489 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
490 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
492 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
493 * - Don't rerun crunchers that only depend on the request URL.
494 * - Don't count redirects and CGI requests as "blocked requests".
496 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
497 * Support crunching based on tags created by server-header taggers.
499 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
500 * Make HTTP snippets static (again). Add a Content-Type for those
501 * with content so the browser doesn't guess it based on the URL.
503 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
504 * Move crunching and request line rebuilding out of chat().
506 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
507 * Introduce tags as alternative way to specify which
508 * actions apply to a request. At the moment tags can be
509 * created based on client and server headers.
511 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
512 * Don't CLF-log CONNECT requests twice.
514 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
515 * Log the source address for ACL-related connection drops.
517 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
518 * New config option: enforce-blocks.
520 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
521 * - Move null byte check into separate function.
522 * - Don't confuse the client with error pages
523 * if a CONNECT request was already confirmed.
525 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
526 * Speed up NULL byte escaping and only log the complete
527 * NULL byte requests with header debugging enabled.
529 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
530 * Answer requests that contain NULL bytes with
531 * a custom response instead of waiting for more
532 * data until the client eventually hangs up.
534 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
535 * - Move delivery and logging of crunched responses
536 * from chat() into send_crunch_response().
537 * - Display the reason for generating http_responses.
538 * - Log the content length for LOG_LEVEL_CLF correctly
539 * (still incorrect for some fixed responses).
540 * - Reword an incorrect comment about
541 * treat-forbidden-connects-like-blocks violating
543 * - Add some log messages.
545 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
546 * Move mutex initialization into separate
547 * function and exit in case of errors.
549 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
550 * - Start to reduce chat()'s line count and move
551 * parts of it into separate functions.
552 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
554 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
555 * - Add Proxy-Agent header to HTTP snippets that are
556 * supposed to reach HTTP clients only.
557 * - Made a few CONNECT log messages more descriptive.
558 * - Catch completely empty server responses (as seen
559 * with Tor's fake ".noconnect" top level domain).
560 * - Use shiny new "forwarding-failed" template for socks errors.
562 * Revision 1.118 2007/01/07 07:43:43 joergs
563 * AmigaOS4 support added.
565 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
566 * Added config option accept-intercepted-requests
567 * and disabled it by default.
569 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
570 * Reverted parts of my last commit
571 * to keep error handling working.
573 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
574 * Fixed gcc43 conversion warnings.
576 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
577 * Fix -pedantic ISO C warning about converting
578 * from function pointer to object pointer.
580 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
581 * Silence compiler warning I introduced with my last commit.
583 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
584 * Mutex protect rand() if POSIX threading
585 * is used, warn the user if that's not possible
586 * and stop using it on _WIN32 where it could
589 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
590 * Don't prevent core dumps by catching SIGABRT.
591 * It's rude and makes debugging unreasonable painful.
593 * Revision 1.110 2006/12/13 14:52:53 etresoft
594 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
596 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
597 * Privoxy is now able to run as intercepting
598 * proxy in combination with any packet filter
599 * that does the port redirection. The destination
600 * is extracted from the "Host:" header which
601 * should be available for nearly all requests.
603 * Moved HTTP snipplets into jcc.c.
604 * Added error message for gopher proxy requests.
606 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
607 * Only unlink the pidfile if it's actually used.
609 * Change order of interception checks to make
610 * it possible to block or redirect requests for
613 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
614 * Make pthread mutex locking more generic. Instead of
615 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
616 * and use mutex locking unless there is an _r function
617 * available. Better safe than sorry.
619 * Fixes "./configure --disable-pthread" and should result
620 * in less threading-related problems on pthread-using platforms,
621 * but it still doesn't fix BR#1122404.
623 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
624 * Move pthread.h inclusion from jcc.c to jcc.h.
625 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
627 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
628 * Don't exit after receiving the second SIGHUP on Solaris.
630 * Fixes BR 1052235, but the same problem may exist on other
631 * systems. Once 3.0.6 is out we should use sigset()
632 * where available and see if it breaks anything.
634 * Revision 1.104 2006/09/23 13:26:38 roro
635 * Replace TABs by spaces in source code.
637 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
638 * Fix +redirect{}. Didn't work with -fast-redirects.
640 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
641 * Respond with 400 and a short text message
642 * if the client tries to use Privoxy as FTP proxy.
644 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
645 * Make number of retries in case of forwarded-connect problems
646 * a config file option (forwarded-connect-retries) and use 0 as
649 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
650 * Set random(3) seed.
652 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
653 * Follow the OpenBSD port's lead and protect the resolve
654 * functions on OpenBSD as well.
656 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
657 * --user fix. Only use the user as group if no group is specified.
658 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
660 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
661 * Windows service (re-)integration
665 * --install[:service_name]
666 * --uninstall[:service_name]
669 * They work as follows:
670 * --install will create a service for you and then terminate.
671 * By default the service name will be "privoxy" (without the quotes).
672 * However you can run multiple services if you wish, just by adding
673 * a colon and then a name (no spaces).
675 * --uninstall follows the exact same rules a --install.
677 * --service is used when the program is executed by the service
678 * control manager, and in normal circumstances would never be
679 * used as a command line argument.
681 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
682 * Windows service integration
684 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
685 * Incorporate Fabian Keil's patch work:
686 http://www.fabiankeil.de/sourcecode/privoxy/
688 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
689 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
690 * with what was really the latest development (the v_3_0_branch branch)
692 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
693 * Thanks to Jindrich Makovicka for a race condition fix for the log
694 * file. The race condition remains for non-pthread implementations.
695 * Reference patch #1175720.
697 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
698 * Add the ability to check jpeg images for invalid
699 * lengths of comment blocks. Defensive strategy
700 * against the exploit:
701 * Microsoft Security Bulletin MS04-028
702 * Buffer Overrun in JPEG Processing (GDI+) Could
703 * Allow Code Execution (833987)
704 * Enabled with +inspect-jpegs in actions files.
706 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
707 * - Fixed usage info for non-unix platforms
708 * - Fixed small cmdline parsing bug
710 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
711 * Diagnostics: Now preserve the returncode of pthread_create
712 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
714 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
715 * No longer ignore SIGCHLD. Fixes bug #769381
717 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
718 * Close jarfile on graceful exit, remove stray line
720 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
721 * Cosmetics: Killed a warning, a typo and an allocation left at exit
723 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
724 * No longer rely on non-POSIX.1 extensions of getcwd().
727 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
728 * Replaced setenv() by posix-compliant putenv()
729 * Thanks to Neil McCalden (nmcc AT users.sf.net).
731 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
732 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
734 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
735 * Clean-up and extension of improvements for forked mode:
736 * - Child's return code now consists of flags RC_FLAG_*
737 * - Reporting toggle to parent now properly #ifdef'ed
738 * - Children now report blocking to parent. This enables
739 * statistics in forked mode
741 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
742 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
743 * Children now report having been toggled through _exit(17), parents
744 * watch for that code and toggle themselves if found.
746 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
747 * Wrapping all *_r functions (the non-_r versions of them) with
748 * mutex semaphores for OSX. Hopefully this will take care of all
749 * of those pesky crash reports.
751 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
752 * Fixed two mostly harmless mem leaks
754 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
755 * Fix: Head of global clients list now initialized to NULL
757 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
758 * Added basic support for OPTIONS and TRACE HTTP methods:
759 * - New interceptor direct_response() added in chat().
760 * - sed() moved to earlier in the process, so that the
761 * Host: header is evaluated before actions and forwarding
764 * Revision 1.92 2002/05/08 16:00:46 oes
765 * Chat's buffer handling:
766 * - Fixed bug with unchecked out-of-mem conditions
767 * while reading client request & server headers
768 * - No longer predict if the buffer limit will be exceeded
769 * in the next read -- check add_to_iob's new
770 * return code. If buffer couldn't be extended
771 * (policy or out-of-mem) while
772 * - reading from client: abort
773 * - reading server headers: send error page
774 * - buffering server body for filter: flush,
775 * and if that fails: send error page
777 * Revision 1.91 2002/04/08 20:35:58 swa
780 * Revision 1.90 2002/04/02 14:57:28 oes
781 * Made sending wafers independent of FEATURE_COOKIE_JAR
783 * Revision 1.89 2002/03/31 17:18:59 jongfoster
784 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
786 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
787 * More compiler warning message maintenance
789 * Revision 1.87 2002/03/26 22:29:54 swa
790 * we have a new homepage!
792 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
793 * Workaround for closing the jarfile before load_config() comes around again
795 * Revision 1.85 2002/03/24 15:23:33 jongfoster
798 * Revision 1.84 2002/03/24 13:25:43 swa
799 * name change related issues
801 * Revision 1.83 2002/03/16 23:54:06 jongfoster
802 * Adding graceful termination feature, to help look for memory leaks.
803 * If you enable this (which, by design, has to be done by hand
804 * editing config.h) and then go to http://i.j.b/die, then the program
805 * will exit cleanly after the *next* request. It should free all the
806 * memory that was used.
808 * Revision 1.82 2002/03/13 00:27:05 jongfoster
811 * Revision 1.81 2002/03/12 01:42:50 oes
812 * Introduced modular filters
814 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
815 * OS/2 port maintenance:
816 * - Fixed EMX build - it had decayed a little
817 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
818 * substituted a memset for now.
820 * Revision 1.79 2002/03/09 20:03:52 jongfoster
821 * - Making various functions return int rather than size_t.
822 * (Undoing a recent change). Since size_t is unsigned on
823 * Windows, functions like read_socket that return -1 on
824 * error cannot return a size_t.
826 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
827 * crashes, and also frequently caused JB to jump to 100%
828 * CPU and stay there. (Because it thought it had just
829 * read ((unsigned)-1) == 4Gb of data...)
831 * - The signature of write_socket has changed, it now simply
832 * returns success=0/failure=nonzero.
834 * - Trying to get rid of a few warnings --with-debug on
835 * Windows, I've introduced a new type "jb_socket". This is
836 * used for the socket file descriptors. On Windows, this
837 * is SOCKET (a typedef for unsigned). Everywhere else, it's
838 * an int. The error value can't be -1 any more, so it's
839 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
840 * Windows it maps to the #define INVALID_SOCKET.)
842 * - The signature of bind_port has changed.
844 * Revision 1.78 2002/03/08 21:35:04 oes
845 * Added optional group supplement to --user option. Will now use default group of user if no group given
847 * Revision 1.77 2002/03/07 03:52:06 oes
848 * - Fixed compiler warnings etc
849 * - Improved handling of failed DNS lookups
851 * Revision 1.76 2002/03/06 22:54:35 jongfoster
852 * Automated function-comment nitpicking.
854 * Revision 1.75 2002/03/06 10:02:19 oes
855 * Fixed stupid bug when --user was not given
857 * Revision 1.74 2002/03/06 00:49:31 jongfoster
858 * Fixing warning on Windows
859 * Making #ifdefs that refer to the same variable consistently
860 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
862 * Revision 1.73 2002/03/05 23:57:30 hal9
863 * Stray character 's' on line 1618 was breaking build.
865 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
866 * - Re-enable OS/2 building after new parms were added
867 * - Fix false out of memory report when resolving CGI templates when no IP
868 * address is available of failed attempt (a la no such domain)
870 * Revision 1.71 2002/03/05 18:13:56 oes
871 * Added --user option
873 * Revision 1.70 2002/03/05 04:52:42 oes
874 * Deleted non-errlog debugging code
876 * Revision 1.69 2002/03/04 23:50:00 jongfoster
877 * Splitting off bind_port() call into bind_port_helper(), with
880 * Revision 1.68 2002/03/04 20:17:32 oes
883 * Revision 1.67 2002/03/04 18:18:57 oes
884 * - Removed _DEBUG mode
885 * - Cleand up cmdline parsing
886 * - Introduced --no-daemon, --pidfile options
887 * - Cleaned up signal handling:
888 * - Terminate cleanly on INT, TERM and ABRT
889 * - Schedule logfile for re-opening on HUP
890 * - Ignore CHLD and PIPE
891 * - Leave the rest with their default handlers
892 * - Uniform handler registration
893 * - Added usage() function
894 * - Played styleguide police
896 * Revision 1.66 2002/03/03 15:06:55 oes
897 * Re-enabled automatic config reloading
899 * Revision 1.65 2002/03/03 14:49:11 oes
900 * Fixed CLF logging: Now uses client's original HTTP request
902 * Revision 1.64 2002/03/03 09:18:03 joergs
903 * Made jumbjuster work on AmigaOS again.
905 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
906 * Clean up a little CRLF unpleasantness that suddenly appeared
908 * Revision 1.62 2002/02/20 23:17:23 jongfoster
909 * Detecting some out-of memory conditions and exiting with a log message.
911 * Revision 1.61 2002/01/17 21:01:52 jongfoster
912 * Moving all our URL and URL pattern parsing code to urlmatch.c.
914 * Revision 1.60 2001/12/30 14:07:32 steudten
915 * - Add signal handling (unix)
916 * - Add SIGHUP handler (unix)
917 * - Add creation of pidfile (unix)
918 * - Add action 'top' in rc file (RH)
919 * - Add entry 'SIGNALS' to manpage
920 * - Add exit message to logfile (unix)
922 * Revision 1.59 2001/12/13 14:07:18 oes
923 * Fixed Bug: 503 error page now sent OK
925 * Revision 1.58 2001/11/30 23:37:24 jongfoster
926 * Renaming the Win32 config file to config.txt - this is almost the
927 * same as the corresponding UNIX name "config"
929 * Revision 1.57 2001/11/16 00:47:43 jongfoster
930 * Changing the tty-disconnection code to use setsid().
932 * Revision 1.56 2001/11/13 20:20:54 jongfoster
933 * Tabs->spaces, fixing a bug with missing {} around an if()
935 * Revision 1.55 2001/11/13 20:14:53 jongfoster
936 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
938 * Revision 1.54 2001/11/07 00:03:14 steudten
939 * Give reliable return value if an error
940 * occurs not just 0 with new daemon mode.
942 * Revision 1.53 2001/11/05 21:41:43 steudten
943 * Add changes to be a real daemon just for unix os.
944 * (change cwd to /, detach from controlling tty, set
945 * process group and session leader to the own process.
947 * Add some fatal-error log message for failed malloc().
948 * Add '-d' if compiled with 'configure --with-debug' to
949 * enable debug output.
951 * Revision 1.52 2001/10/26 20:11:20 jongfoster
952 * Fixing type mismatch
954 * Revision 1.51 2001/10/26 17:38:28 oes
957 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
958 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
959 * threads to call select() simultaneously. So, it's time to do a real, live,
960 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
961 * (native). Both versions will work, but using __OS2__ offers multi-threading.
963 * Revision 1.49 2001/10/23 21:41:35 jongfoster
964 * Added call to initialize the (statically-allocated of course)
965 * "out of memory" CGI response.
967 * Revision 1.48 2001/10/10 19:56:46 jongfoster
968 * Moving some code that wasn't cookie-related out of an #ifdef
971 * Revision 1.47 2001/10/10 16:44:36 oes
972 * Added CONNECT destination port limitation check
974 * Revision 1.46 2001/10/08 15:17:41 oes
975 * Re-enabled SSL forwarding
977 * Revision 1.45 2001/10/07 15:42:11 oes
978 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
980 * Moved downgrading of the HTTP version from parse_http_request to
981 * chat(), since we can't decide if it is necessary before we have
982 * determined the actions for the URL. The HTTP command is now
983 * *always* re-built so the repairs need no longer be special-cased.
985 * filter_popups now gets a csp pointer so it can raise the new
986 * CSP_FLAG_MODIFIED flag.
990 * Added configurable size limit for the IOB. If the IOB grows so
991 * large that the next read would exceed the limit, the header
992 * is generated, and the header & unfiltered buffer are flushed
993 * to the client. Chat then continues in non-buffering,
994 * non-filtering body mode.
996 * Revision 1.44 2001/10/02 18:13:57 oes
999 * Revision 1.43 2001/10/02 15:32:13 oes
1000 * Moved generation of hdr
1002 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
1003 * Cleaning up 2 compiler warnings on OS/2.
1005 * Revision 1.41 2001/09/16 17:05:14 jongfoster
1006 * Removing unused #include showarg.h
1008 * Revision 1.40 2001/09/16 15:41:45 jongfoster
1009 * Fixing signed/unsigned comparison warning.
1011 * Revision 1.39 2001/09/16 13:21:27 jongfoster
1012 * Changes to use new list functions.
1014 * Revision 1.38 2001/09/16 13:01:46 jongfoster
1015 * Removing redundant function call that zeroed zalloc()'d memory.
1017 * Revision 1.37 2001/09/10 11:12:24 oes
1018 * Deleted unused variable
1020 * Revision 1.36 2001/09/10 10:56:15 oes
1021 * Silenced compiler warnings
1023 * Revision 1.35 2001/07/31 14:44:22 oes
1024 * Deleted unused size parameter from filter_popups()
1026 * Revision 1.34 2001/07/30 22:08:36 jongfoster
1027 * Tidying up #defines:
1028 * - All feature #defines are now of the form FEATURE_xxx
1029 * - Permanently turned off WIN_GUI_EDIT
1030 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
1032 * Revision 1.33 2001/07/29 19:32:00 jongfoster
1033 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
1035 * Revision 1.32 2001/07/29 18:47:05 jongfoster
1036 * Adding missing #include "loadcfg.h"
1038 * Revision 1.31 2001/07/29 12:17:48 oes
1039 * Applied pthread fix by Paul Lieverse
1041 * Revision 1.30 2001/07/25 22:57:13 jongfoster
1042 * __BEOS__ no longer overrides FEATURE_PTHREAD.
1043 * This is because FEATURE_PTHREAD will soon be widely used, so I
1044 * want to keep it simple.
1046 * Revision 1.29 2001/07/24 12:47:06 oes
1047 * Applied BeOS support update by Eugenia
1049 * Revision 1.28 2001/07/23 13:26:12 oes
1050 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
1052 * Revision 1.27 2001/07/19 19:09:47 haroon
1053 * - Added code to take care of the situation where while processing the first
1054 * server response (which includes the server header), after finding the end
1055 * of the headers we were not looking past the end of the headers for
1056 * content modification. I enabled it for filter_popups.
1057 * Someone else should look to see if other similar operations should be
1058 * done to the discarded portion of the buffer.
1060 * Note 2001/07/20: No, the other content modification mechanisms will process
1061 * the whole iob later anyway. --oes
1063 * Revision 1.26 2001/07/18 12:31:36 oes
1066 * Revision 1.25 2001/07/15 19:43:49 jongfoster
1067 * Supports POSIX threads.
1068 * Also removed some unused #includes.
1070 * Revision 1.24 2001/07/13 14:00:40 oes
1071 * - Generic content modification scheme:
1072 * Each feature has its own applicability flag that is set
1073 * from csp->action->flags.
1074 * Replaced the "filtering" int flag , by a function pointer
1075 * "content_filter" to the function that will do the content
1076 * modification. If it is != NULL, the document will be buffered
1077 * and processed through *content_filter, which must set
1078 * csp->content_length and return a modified copy of the body
1079 * or return NULL (on failiure).
1080 * - Changed csp->is_text to the more generic bitmap csp->content_type
1081 * which can currently take the valued CT_TEXT or CT_GIF
1082 * - Reformatting etc
1083 * - Removed all #ifdef PCRS
1085 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
1086 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
1087 * the `block_acl' call. This prevents a compilation error when the user
1088 * does not wish to use the "ACL" feature.
1090 * Revision 1.22 2001/06/29 21:45:41 oes
1091 * Indentation, CRLF->LF, Tab-> Space
1093 * Revision 1.21 2001/06/29 13:29:36 oes
1094 * - Cleaned up, improved comments
1095 * - Unified all possible interceptors (CGI,
1096 * block, trust, fast_redirect) in one
1097 * place, with one (CGI) answer generation
1098 * mechansim. Much clearer now.
1099 * - Removed the GIF image generation, which
1100 * is now done in filters.c:block_url()
1101 * - Made error conditions like domain lookup
1102 * failiure or (various) problems while talking
1103 * to the server use cgi.c:error_response()
1104 * instead of generating HTML/HTTP in chat() (yuck!)
1105 * - Removed logentry from cancelled commit
1107 * Revision 1.20 2001/06/09 10:55:28 jongfoster
1108 * Changing BUFSIZ ==> BUFFER_SIZE
1110 * Revision 1.19 2001/06/07 23:12:52 jongfoster
1111 * Replacing function pointer in struct gateway with a directly
1112 * called function forwarded_connect().
1113 * Replacing struct gateway with struct forward_spec
1115 * Revision 1.18 2001/06/03 19:12:16 oes
1116 * introduced new cgi handling
1118 * Revision 1.17 2001/06/01 20:07:23 jongfoster
1119 * Now uses action +image-blocker{} rather than config->tinygif
1121 * Revision 1.16 2001/06/01 18:49:17 jongfoster
1122 * Replaced "list_share" with "list" - the tiny memory gain was not
1123 * worth the extra complexity.
1125 * Revision 1.15 2001/05/31 21:24:47 jongfoster
1126 * Changed "permission" to "action" throughout.
1127 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
1128 * Moved vanilla wafer check into chat(), since we must now
1129 * decide whether or not to add it based on the URL.
1131 * Revision 1.14 2001/05/29 20:14:01 joergs
1132 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
1135 * Revision 1.13 2001/05/29 09:50:24 jongfoster
1136 * Unified blocklist/imagelist/permissionslist.
1137 * File format is still under discussion, but the internal changes
1138 * are (mostly) done.
1140 * Also modified interceptor behaviour:
1141 * - We now intercept all URLs beginning with one of the following
1142 * prefixes (and *only* these prefixes):
1144 * * http://ijbswa.sf.net/config/
1145 * * http://ijbswa.sourceforge.net/config/
1146 * - New interceptors "home page" - go to http://i.j.b/ to see it.
1147 * - Internal changes so that intercepted and fast redirect pages
1148 * are not replaced with an image.
1149 * - Interceptors now have the option to send a binary page direct
1150 * to the client. (i.e. ijb-send-banner uses this)
1151 * - Implemented show-url-info interceptor. (Which is why I needed
1152 * the above interceptors changes - a typical URL is
1153 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
1154 * The previous mechanism would not have intercepted that, and
1155 * if it had been intercepted then it then it would have replaced
1156 * it with an image.)
1158 * Revision 1.12 2001/05/27 22:17:04 oes
1160 * - re_process_buffer no longer writes the modified buffer
1161 * to the client, which was very ugly. It now returns the
1162 * buffer, which it is then written by chat.
1164 * - content_length now adjusts the Content-Length: header
1165 * for modified documents rather than crunch()ing it.
1166 * (Length info in csp->content_length, which is 0 for
1167 * unmodified documents)
1169 * - For this to work, sed() is called twice when filtering.
1171 * Revision 1.11 2001/05/26 17:27:53 jongfoster
1172 * Added support for CLF and fixed LOG_LEVEL_LOG.
1173 * Also did CRLF->LF fix of my previous patch.
1175 * Revision 1.10 2001/05/26 15:26:15 jongfoster
1176 * ACL feature now provides more security by immediately dropping
1177 * connections from untrusted hosts.
1179 * Revision 1.9 2001/05/26 00:28:36 jongfoster
1180 * Automatic reloading of config file.
1181 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
1182 * Most of the global variables have been moved to a new
1183 * struct configuration_spec, accessed through csp->config->globalname
1184 * Most of the globals remaining are used by the Win32 GUI.
1186 * Revision 1.8 2001/05/25 22:43:18 jongfoster
1187 * Fixing minor memory leak and buffer overflow.
1189 * Revision 1.7 2001/05/25 22:34:30 jongfoster
1192 * Revision 1.6 2001/05/23 00:13:58 joergs
1193 * AmigaOS support fixed.
1195 * Revision 1.5 2001/05/22 18:46:04 oes
1197 * - Enabled filtering banners by size rather than URL
1198 * by adding patterns that replace all standard banner
1199 * sizes with the "Junkbuster" gif to the re_filterfile
1201 * - Enabled filtering WebBugs by providing a pattern
1202 * which kills all 1x1 images
1204 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
1205 * which is selected by the (nonstandard and therefore
1206 * capital) letter 'U' in the option string.
1207 * It causes the quantifiers to be ungreedy by default.
1208 * Appending a ? turns back to greedy (!).
1210 * - Added a new interceptor ijb-send-banner, which
1211 * sends back the "Junkbuster" gif. Without imagelist or
1212 * MSIE detection support, or if tinygif = 1, or the
1213 * URL isn't recognized as an imageurl, a lame HTML
1214 * explanation is sent instead.
1216 * - Added new feature, which permits blocking remote
1217 * script redirects and firing back a local redirect
1219 * The feature is conditionally compiled, i.e. it
1220 * can be disabled with --disable-fast-redirects,
1221 * plus it must be activated by a "fast-redirects"
1222 * line in the config file, has its own log level
1223 * and of course wants to be displayed by show-proxy-args
1224 * Note: Boy, all the #ifdefs in 1001 locations and
1225 * all the fumbling with configure.in and acconfig.h
1226 * were *way* more work than the feature itself :-(
1228 * - Because a generic redirect template was needed for
1229 * this, tinygif = 3 now uses the same.
1231 * - Moved GIFs, and other static HTTP response templates
1234 * - Some minor fixes
1236 * - Removed some >400 CRs again (Jon, you really worked
1239 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1240 * Made failure to bind() a fatal error.
1242 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1243 * Version 2.9.4 checkin.
1244 * - Merged popupfile and cookiefile, and added control over PCRS
1245 * filtering, in new "permissionsfile".
1246 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1247 * file error you now get a message box (in the Win32 GUI) rather
1248 * than the program exiting with no explanation.
1249 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1251 * - Removed tabs from "config"
1252 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1253 * - Bumped up version number.
1255 * Revision 1.2 2001/05/17 22:34:44 oes
1256 * - Added hint on GIF char array generation to jcc.c
1257 * - Cleaned CRLF's from the sources and related files
1258 * - Repaired logging for REF and FRC
1260 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1261 * Initial import of version 2.9.3 source tree
1264 *********************************************************************/
1270 #include <sys/types.h>
1279 # ifndef FEATURE_PTHREAD
1283 # include <windows.h>
1284 # include <process.h>
1285 # endif /* ndef FEATURE_PTHREAD */
1288 # ifndef _WIN_CONSOLE
1289 # include "w32log.h"
1290 # endif /* ndef _WIN_CONSOLE */
1291 # include "w32svrapi.h"
1293 #else /* ifndef _WIN32 */
1295 # if !defined (__OS2__)
1296 # include <unistd.h>
1297 # include <sys/wait.h>
1298 # endif /* ndef __OS2__ */
1299 # include <sys/time.h>
1300 # include <sys/stat.h>
1301 # include <sys/ioctl.h>
1304 #include <sys/termios.h>
1312 # include <signal.h>
1315 # include <socket.h> /* BeOS has select() for sockets only. */
1316 # include <OS.h> /* declarations for threads and stuff. */
1319 # if defined(__EMX__) || defined(__OS2__)
1320 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1325 #define bzero(B,N) memset(B,0x00,n)
1329 # include <select.h>
1334 #include "project.h"
1337 #include "filters.h"
1338 #include "loaders.h"
1339 #include "parsers.h"
1340 #include "miscutil.h"
1342 #include "jbsockets.h"
1343 #include "gateway.h"
1344 #include "actions.h"
1346 #include "loadcfg.h"
1347 #include "urlmatch.h"
1349 const char jcc_h_rcs[] = JCC_H_VERSION;
1350 const char project_h_rcs[] = PROJECT_H_VERSION;
1353 struct client_state clients[1];
1354 struct file_list files[1];
1356 #ifdef FEATURE_STATISTICS
1357 int urls_read = 0; /* total nr of urls read inc rejected */
1358 int urls_rejected = 0; /* total nr of urls rejected */
1359 #endif /* def FEATURE_STATISTICS */
1361 #ifdef FEATURE_GRACEFUL_TERMINATION
1362 int g_terminate = 0;
1365 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1366 static void sig_handler(int the_signal);
1368 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1369 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1370 static jb_err get_server_headers(struct client_state *csp);
1371 static const char *crunch_reason(const struct http_response *rsp);
1372 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1373 static char *get_request_line(struct client_state *csp);
1374 static jb_err receive_client_request(struct client_state *csp);
1375 static jb_err parse_client_request(struct client_state *csp);
1376 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1377 static jb_err change_request_destination(struct client_state *csp);
1378 static void chat(struct client_state *csp);
1379 static void serve(struct client_state *csp);
1380 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1381 static void usage(const char *myname);
1383 static void initialize_mutexes(void);
1384 static jb_socket bind_port_helper(struct configuration_spec *config);
1385 static void listen_loop(void);
1388 void serve(struct client_state *csp);
1389 #else /* ifndef AMIGA */
1390 static void serve(struct client_state *csp);
1391 #endif /* def AMIGA */
1394 static int32 server_thread(void *data);
1395 #endif /* def __BEOS__ */
1398 #define sleep(N) Sleep(((N) * 1000))
1402 #define sleep(N) DosSleep(((N) * 100))
1405 #ifdef MUTEX_LOCKS_AVAILABLE
1407 * XXX: Does the locking stuff really belong in this file?
1409 privoxy_mutex_t log_mutex;
1410 privoxy_mutex_t log_init_mutex;
1411 privoxy_mutex_t connection_reuse_mutex;
1413 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1414 privoxy_mutex_t resolver_mutex;
1415 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1417 #ifndef HAVE_GMTIME_R
1418 privoxy_mutex_t gmtime_mutex;
1419 #endif /* ndef HAVE_GMTIME_R */
1421 #ifndef HAVE_LOCALTIME_R
1422 privoxy_mutex_t localtime_mutex;
1423 #endif /* ndef HAVE_GMTIME_R */
1426 privoxy_mutex_t rand_mutex;
1427 #endif /* ndef HAVE_RANDOM */
1429 #endif /* def MUTEX_LOCKS_AVAILABLE */
1432 const char *basedir = NULL;
1433 const char *pidfile = NULL;
1434 static int received_hup_signal = 0;
1435 #endif /* defined unix */
1437 /* HTTP snipplets. */
1438 static const char CSUCCEED[] =
1439 "HTTP/1.0 200 Connection established\r\n"
1440 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1442 static const char CHEADER[] =
1443 "HTTP/1.0 400 Invalid header received from client\r\n"
1444 "Proxy-Agent: Privoxy " VERSION "\r\n"
1445 "Content-Type: text/plain\r\n"
1446 "Connection: close\r\n\r\n"
1447 "Invalid header received from client.\r\n";
1449 static const char FTP_RESPONSE[] =
1450 "HTTP/1.0 400 Invalid request received from client\r\n"
1451 "Content-Type: text/plain\r\n"
1452 "Connection: close\r\n\r\n"
1453 "Invalid request. Privoxy doesn't support FTP.\r\n";
1455 static const char GOPHER_RESPONSE[] =
1456 "HTTP/1.0 400 Invalid request received from client\r\n"
1457 "Content-Type: text/plain\r\n"
1458 "Connection: close\r\n\r\n"
1459 "Invalid request. Privoxy doesn't support gopher.\r\n";
1461 /* XXX: should be a template */
1462 static const char MISSING_DESTINATION_RESPONSE[] =
1463 "HTTP/1.0 400 Bad request received from client\r\n"
1464 "Proxy-Agent: Privoxy " VERSION "\r\n"
1465 "Content-Type: text/plain\r\n"
1466 "Connection: close\r\n\r\n"
1467 "Bad request. Privoxy was unable to extract the destination.\r\n";
1469 /* XXX: should be a template */
1470 static const char NO_SERVER_DATA_RESPONSE[] =
1471 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1472 "Proxy-Agent: Privoxy " VERSION "\r\n"
1473 "Content-Type: text/plain\r\n"
1474 "Connection: close\r\n\r\n"
1475 "Empty server or forwarder response.\r\n"
1476 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1478 /* XXX: should be a template */
1479 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1480 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1481 "Proxy-Agent: Privoxy " VERSION "\r\n"
1482 "Content-Type: text/plain\r\n"
1483 "Connection: close\r\n\r\n"
1484 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1487 /* XXX: should be a template */
1488 static const char NULL_BYTE_RESPONSE[] =
1489 "HTTP/1.0 400 Bad request received from client\r\n"
1490 "Proxy-Agent: Privoxy " VERSION "\r\n"
1491 "Content-Type: text/plain\r\n"
1492 "Connection: close\r\n\r\n"
1493 "Bad request. Null byte(s) before end of request.\r\n";
1496 /* XXX: should be a template */
1497 static const char MESSED_UP_REQUEST_RESPONSE[] =
1498 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1499 "Proxy-Agent: Privoxy " VERSION "\r\n"
1500 "Content-Type: text/plain\r\n"
1501 "Connection: close\r\n\r\n"
1502 "Bad request. Messed up with header filters.\r\n";
1504 static const char TOO_MANY_CONNECTIONS_RESPONSE[] =
1505 "HTTP/1.0 503 Too many open connections\r\n"
1506 "Proxy-Agent: Privoxy " VERSION "\r\n"
1507 "Content-Type: text/plain\r\n"
1508 "Connection: close\r\n\r\n"
1509 "Maximum number of open connections reached.\r\n";
1511 /* XXX: should be a template */
1512 static const char CONNECTION_TIMEOUT_RESPONSE[] =
1513 "HTTP/1.0 504 Connection timeout\r\n"
1514 "Proxy-Agent: Privoxy " VERSION "\r\n"
1515 "Content-Type: text/plain\r\n"
1516 "Connection: close\r\n\r\n"
1517 "The connection timed out.\r\n";
1519 /* A function to crunch a response */
1520 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1522 /* Crunch function flags */
1523 #define CF_NO_FLAGS 0
1524 /* Cruncher applies to forced requests as well */
1525 #define CF_IGNORE_FORCE 1
1526 /* Crunched requests are counted for the block statistics */
1527 #define CF_COUNT_AS_REJECT 2
1529 /* A crunch function and its flags */
1532 const crunch_func_ptr cruncher;
1536 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1538 /* Complete list of cruncher functions */
1539 static const struct cruncher crunchers_all[] = {
1540 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1541 { block_url, CF_COUNT_AS_REJECT },
1542 #ifdef FEATURE_TRUST
1543 { trust_url, CF_COUNT_AS_REJECT },
1544 #endif /* def FEATURE_TRUST */
1545 { redirect_url, CF_NO_FLAGS },
1546 { dispatch_cgi, CF_IGNORE_FORCE},
1550 /* Light version, used after tags are applied */
1551 static const struct cruncher crunchers_light[] = {
1552 { block_url, CF_COUNT_AS_REJECT },
1553 { redirect_url, CF_NO_FLAGS },
1559 * XXX: Don't we really mean
1565 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1566 /*********************************************************************
1568 * Function : sig_handler
1570 * Description : Signal handler for different signals.
1571 * Exit gracefully on TERM and INT
1572 * or set a flag that will cause the errlog
1573 * to be reopened by the main thread on HUP.
1576 * 1 : the_signal = the signal cause this function to call
1580 *********************************************************************/
1581 static void sig_handler(int the_signal)
1587 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1599 received_hup_signal = 1;
1605 * We shouldn't be here, unless we catch signals
1606 * in main() that we can't handle here!
1608 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1616 /*********************************************************************
1618 * Function : client_protocol_is_unsupported
1620 * Description : Checks if the client used a known unsupported
1621 * protocol and deals with it by sending an error
1625 * 1 : csp = Current client state (buffers, headers, etc...)
1626 * 2 : req = the first request line send by the client
1628 * Returns : TRUE if an error response has been generated, or
1629 * FALSE if the request doesn't look invalid.
1631 *********************************************************************/
1632 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1635 * If it's a FTP or gopher request, we don't support it.
1637 * These checks are better than nothing, but they might
1638 * not work in all configurations and some clients might
1639 * have problems digesting the answer.
1641 * They should, however, never cause more problems than
1642 * Privoxy's old behaviour (returning the misleading HTML
1645 * "Could not resolve http://(ftp|gopher)://example.org").
1647 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1649 const char *response = NULL;
1650 const char *protocol = NULL;
1652 if (!strncmpic(req, "GET ftp://", 10))
1654 response = FTP_RESPONSE;
1659 response = GOPHER_RESPONSE;
1660 protocol = "GOPHER";
1662 log_error(LOG_LEVEL_ERROR,
1663 "%s tried to use Privoxy as %s proxy: %s",
1664 csp->ip_addr_str, protocol, req);
1665 log_error(LOG_LEVEL_CLF,
1666 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1668 write_socket(csp->cfd, response, strlen(response));
1677 /*********************************************************************
1679 * Function : get_request_destination_elsewhere
1681 * Description : If the client's request was redirected into
1682 * Privoxy without the client's knowledge,
1683 * the request line lacks the destination host.
1685 * This function tries to get it elsewhere,
1686 * provided accept-intercepted-requests is enabled.
1688 * "Elsewhere" currently only means "Host: header",
1689 * but in the future we may ask the redirecting
1690 * packet filter to look the destination up.
1692 * If the destination stays unknown, an error
1693 * response is send to the client and headers
1694 * are freed so that chat() can return directly.
1697 * 1 : csp = Current client state (buffers, headers, etc...)
1698 * 2 : headers = a header list
1700 * Returns : JB_ERR_OK if the destination is now known, or
1701 * JB_ERR_PARSE if it isn't.
1703 *********************************************************************/
1704 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1708 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1710 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1711 " Privoxy isn't configured to accept intercepted requests.",
1712 csp->ip_addr_str, csp->http->cmd);
1713 /* XXX: Use correct size */
1714 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1715 csp->ip_addr_str, csp->http->cmd);
1717 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1718 destroy_list(headers);
1720 return JB_ERR_PARSE;
1722 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1724 /* Split the domain we just got for pattern matching */
1725 init_domain_components(csp->http);
1731 /* We can't work without destination. Go spread the news.*/
1733 req = list_to_text(headers);
1735 /* XXX: Use correct size */
1736 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1737 csp->ip_addr_str, csp->http->cmd);
1738 log_error(LOG_LEVEL_ERROR,
1739 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1740 csp->ip_addr_str, csp->http->cmd, req);
1743 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1744 destroy_list(headers);
1746 return JB_ERR_PARSE;
1749 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1750 * to get the destination IP address, use it as host directly
1751 * or do a reverse DNS lookup first.
1756 /*********************************************************************
1758 * Function : get_server_headers
1760 * Description : Parses server headers in iob and fills them
1761 * into csp->headers so that they can later be
1765 * 1 : csp = Current client state (buffers, headers, etc...)
1767 * Returns : JB_ERR_OK if everything went fine, or
1768 * JB_ERR_PARSE if the headers were incomplete.
1770 *********************************************************************/
1771 static jb_err get_server_headers(struct client_state *csp)
1773 int continue_hack_in_da_house = 0;
1776 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1781 * continue hack in da house. Ignore the ending of
1782 * this head and continue enlisting header lines.
1783 * The reason is described below.
1785 enlist(csp->headers, "");
1786 continue_hack_in_da_house = 0;
1789 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1792 * It's a bodyless continue response, don't
1793 * stop header parsing after reaching its end.
1795 * As a result Privoxy will concatenate the
1796 * next response's head and parse and deliver
1797 * the headers as if they belonged to one request.
1799 * The client will separate them because of the
1800 * empty line between them.
1802 * XXX: What we're doing here is clearly against
1803 * the intended purpose of the continue header,
1804 * and under some conditions (HTTP/1.0 client request)
1805 * it's a standard violation.
1807 * Anyway, "sort of against the spec" is preferable
1808 * to "always getting confused by Continue responses"
1809 * (Privoxy's behaviour before this hack was added)
1811 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1812 continue_hack_in_da_house = 1;
1814 else if (*header == '\0')
1817 * If the header is empty, but the Continue hack
1818 * isn't active, we can assume that we reached the
1819 * end of the buffer before we hit the end of the
1822 * Inform the caller an let it decide how to handle it.
1824 return JB_ERR_PARSE;
1827 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1830 * XXX: Should we quit the request and return a
1831 * out of memory error page instead?
1833 log_error(LOG_LEVEL_ERROR,
1834 "Out of memory while enlisting server headers. %s lost.",
1844 /*********************************************************************
1846 * Function : crunch_reason
1848 * Description : Translates the crunch reason code into a string.
1851 * 1 : rsp = a http_response
1853 * Returns : A string with the crunch reason or an error description.
1855 *********************************************************************/
1856 static const char *crunch_reason(const struct http_response *rsp)
1858 char * reason = NULL;
1860 assert(rsp != NULL);
1863 return "Internal error while searching for crunch reason";
1866 switch (rsp->reason)
1868 case RSP_REASON_UNSUPPORTED:
1869 reason = "Unsupported HTTP feature";
1871 case RSP_REASON_BLOCKED:
1874 case RSP_REASON_UNTRUSTED:
1875 reason = "Untrusted";
1877 case RSP_REASON_REDIRECTED:
1878 reason = "Redirected";
1880 case RSP_REASON_CGI_CALL:
1881 reason = "CGI Call";
1883 case RSP_REASON_NO_SUCH_DOMAIN:
1884 reason = "DNS failure";
1886 case RSP_REASON_FORWARDING_FAILED:
1887 reason = "Forwarding failed";
1889 case RSP_REASON_CONNECT_FAILED:
1890 reason = "Connection failure";
1892 case RSP_REASON_OUT_OF_MEMORY:
1893 reason = "Out of memory (may mask other reasons)";
1896 reason = "No reason recorded";
1904 /*********************************************************************
1906 * Function : send_crunch_response
1908 * Description : Delivers already prepared response for
1909 * intercepted requests, logs the interception
1910 * and frees the response.
1913 * 1 : csp = Current client state (buffers, headers, etc...)
1914 * 1 : rsp = Fully prepared response. Will be freed on exit.
1916 * Returns : Nothing.
1918 *********************************************************************/
1919 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1921 const struct http_request *http = csp->http;
1922 char status_code[4];
1924 assert(rsp != NULL);
1925 assert(rsp->head != NULL);
1930 * Not supposed to happen. If it does
1931 * anyway, treat it as an unknown error.
1933 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1934 /* return code doesn't matter */
1939 /* If rsp is still NULL, we have serious internal problems. */
1940 log_error(LOG_LEVEL_FATAL,
1941 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1945 * Extract the status code from the actual head
1946 * that was send to the client. It is the only
1947 * way to get it right for all requests, including
1948 * the fixed ones for out-of-memory problems.
1950 * A head starts like this: 'HTTP/1.1 200...'
1954 status_code[0] = rsp->head[9];
1955 status_code[1] = rsp->head[10];
1956 status_code[2] = rsp->head[11];
1957 status_code[3] = '\0';
1959 /* Write the answer to the client */
1960 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1961 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1963 /* There is nothing we can do about it. */
1964 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1967 /* Log that the request was crunched and why. */
1968 log_error(LOG_LEVEL_CRUNCH, "%s: %s", crunch_reason(rsp), http->url);
1969 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %u",
1970 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1972 /* Clean up and return */
1973 if (cgi_error_memory() != rsp)
1975 free_http_response(rsp);
1982 /*********************************************************************
1984 * Function : request_contains_null_bytes
1986 * Description : Checks for NULL bytes in the request and sends
1987 * an error message to the client if any were found.
1989 * XXX: currently not used, see comment in chat().
1992 * 1 : csp = Current client state (buffers, headers, etc...)
1993 * 2 : buf = Data from the client's request to check.
1994 * 3 : len = The data length.
1996 * Returns : TRUE if the request contained one or more NULL bytes, or
1999 *********************************************************************/
2000 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
2002 size_t c_len; /* Request lenght when treated as C string */
2004 c_len = strlen(buf);
2009 * Null byte(s) found. Log the request,
2010 * return an error response and hang up.
2012 size_t tmp_len = c_len;
2017 * Replace NULL byte(s) with '°' characters
2018 * so the request can be logged as string.
2019 * XXX: Is there a better replacement character?
2022 tmp_len += strlen(buf+tmp_len);
2023 } while (tmp_len < len);
2025 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
2026 "(length=%d, strlen=%u).", csp->ip_addr_str, len, c_len);
2027 log_error(LOG_LEVEL_HEADER,
2028 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
2030 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
2032 /* XXX: Log correct size */
2033 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2043 /*********************************************************************
2045 * Function : crunch_response_triggered
2047 * Description : Checks if the request has to be crunched,
2048 * and delivers the crunch response if necessary.
2051 * 1 : csp = Current client state (buffers, headers, etc...)
2052 * 2 : crunchers = list of cruncher functions to run
2054 * Returns : TRUE if the request was answered with a crunch response
2057 *********************************************************************/
2058 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
2060 struct http_response *rsp = NULL;
2061 const struct cruncher *c;
2064 * If CGI request crunching is disabled,
2065 * check the CGI dispatcher out of order to
2066 * prevent unintentional blocks or redirects.
2068 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
2069 && (NULL != (rsp = dispatch_cgi(csp))))
2071 /* Deliver, log and free the interception response. */
2072 send_crunch_response(csp, rsp);
2076 for (c = crunchers; c->cruncher != NULL; c++)
2079 * Check the cruncher if either Privoxy is toggled
2080 * on and the request isn't forced, or if the cruncher
2081 * applies to forced requests as well.
2083 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
2084 !(csp->flags & CSP_FLAG_FORCED)) ||
2085 (c->flags & CF_IGNORE_FORCE))
2087 rsp = c->cruncher(csp);
2090 /* Deliver, log and free the interception response. */
2091 send_crunch_response(csp, rsp);
2092 #ifdef FEATURE_STATISTICS
2093 if (c->flags & CF_COUNT_AS_REJECT)
2095 csp->flags |= CSP_FLAG_REJECTED;
2097 #endif /* def FEATURE_STATISTICS */
2108 /*********************************************************************
2110 * Function : build_request_line
2112 * Description : Builds the HTTP request line.
2114 * If a HTTP forwarder is used it expects the whole URL,
2115 * web servers only get the path.
2118 * 1 : csp = Current client state (buffers, headers, etc...)
2119 * 2 : fwd = The forwarding spec used for the request
2120 * XXX: Should use http->fwd instead.
2121 * 3 : request_line = The old request line which will be replaced.
2123 * Returns : Nothing. Terminates in case of memory problems.
2125 *********************************************************************/
2126 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
2128 struct http_request *http = csp->http;
2130 assert(http->ssl == 0);
2133 * Downgrade http version from 1.1 to 1.0
2134 * if +downgrade action applies.
2136 if ( (csp->action->flags & ACTION_DOWNGRADE)
2137 && (!strcmpic(http->ver, "HTTP/1.1")))
2140 http->ver = strdup("HTTP/1.0");
2142 if (http->ver == NULL)
2144 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
2149 * Rebuild the request line.
2151 freez(*request_line);
2152 *request_line = strdup(http->gpc);
2153 string_append(request_line, " ");
2155 if (fwd->forward_host)
2157 string_append(request_line, http->url);
2161 string_append(request_line, http->path);
2163 string_append(request_line, " ");
2164 string_append(request_line, http->ver);
2166 if (*request_line == NULL)
2168 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
2170 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
2174 /*********************************************************************
2176 * Function : change_request_destination
2178 * Description : Parse a (rewritten) request line and regenerate
2179 * the http request data.
2182 * 1 : csp = Current client state (buffers, headers, etc...)
2184 * Returns : Forwards the parse_http_request() return code.
2185 * Terminates in case of memory problems.
2187 *********************************************************************/
2188 static jb_err change_request_destination(struct client_state *csp)
2190 struct http_request *http = csp->http;
2193 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
2194 free_http_request(http);
2195 err = parse_http_request(csp->headers->first->str, http);
2196 if (JB_ERR_OK != err)
2198 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
2199 jb_err_to_string(err));
2203 /* XXX: ocmd is a misleading name */
2204 http->ocmd = strdup(http->cmd);
2205 if (http->ocmd == NULL)
2207 log_error(LOG_LEVEL_FATAL,
2208 "Out of memory copying rewritten HTTP request line");
2216 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2217 /*********************************************************************
2219 * Function : server_response_is_complete
2221 * Description : Determines whether we should stop reading
2222 * from the server socket.
2225 * 1 : csp = Current client state (buffers, headers, etc...)
2226 * 2 : content_length = Length of content received so far.
2228 * Returns : TRUE if the response is complete,
2231 *********************************************************************/
2232 static int server_response_is_complete(struct client_state *csp,
2233 unsigned long long content_length)
2235 int content_length_known = !!(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2237 if (!strcmpic(csp->http->gpc, "HEAD"))
2240 * "HEAD" implies no body, we are thus expecting
2241 * no content. XXX: incomplete "list" of methods?
2243 csp->expected_content_length = 0;
2244 content_length_known = TRUE;
2247 if (csp->http->status == 304)
2250 * Expect no body. XXX: incomplete "list" of status codes?
2252 csp->expected_content_length = 0;
2253 content_length_known = TRUE;
2256 return (content_length_known && ((0 == csp->expected_content_length)
2257 || (csp->expected_content_length <= content_length)));
2261 /*********************************************************************
2263 * Function : wait_for_alive_connections
2265 * Description : Waits for alive connections to timeout.
2271 *********************************************************************/
2272 static void wait_for_alive_connections()
2274 int connections_alive = close_unusable_connections();
2276 while (0 < connections_alive)
2278 log_error(LOG_LEVEL_CONNECT,
2279 "Waiting for %d connections to timeout.",
2282 connections_alive = close_unusable_connections();
2285 log_error(LOG_LEVEL_CONNECT, "No connections to wait for left.");
2288 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2291 /*********************************************************************
2293 * Function : mark_server_socket_tainted
2295 * Description : Makes sure we don't reuse a server socket
2296 * (if we didn't read everything the server sent
2297 * us reusing the socket would lead to garbage).
2300 * 1 : csp = Current client state (buffers, headers, etc...)
2304 *********************************************************************/
2305 static void mark_server_socket_tainted(struct client_state *csp)
2307 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2309 log_error(LOG_LEVEL_CONNECT, "Unsetting keep-alive flag.");
2310 csp->flags &= ~CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE;
2314 /*********************************************************************
2316 * Function : get_request_line
2318 * Description : Read the client request line.
2321 * 1 : csp = Current client state (buffers, headers, etc...)
2323 * Returns : Pointer to request line or NULL in case of errors.
2325 *********************************************************************/
2326 static char *get_request_line(struct client_state *csp)
2328 char buf[BUFFER_SIZE];
2329 char *request_line = NULL;
2332 memset(buf, 0, sizeof(buf));
2336 if (!data_is_available(csp->cfd, csp->config->socket_timeout))
2338 log_error(LOG_LEVEL_ERROR,
2339 "Stopped waiting for the request line.");
2340 write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE,
2341 strlen(CONNECTION_TIMEOUT_RESPONSE));
2345 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2347 if (len <= 0) return NULL;
2350 * If there is no memory left for buffering the
2351 * request, there is nothing we can do but hang up
2353 if (add_to_iob(csp, buf, len))
2358 request_line = get_header(csp->iob);
2360 } while ((NULL != request_line) && ('\0' == *request_line));
2362 return request_line;
2367 /*********************************************************************
2369 * Function : receive_client_request
2371 * Description : Read the client's request (more precisely the
2372 * client headers) and answer it if necessary.
2374 * Note that since we're not using select() we could get
2375 * blocked here if a client connected, then didn't say
2379 * 1 : csp = Current client state (buffers, headers, etc...)
2381 * Returns : JB_ERR_OK, JB_ERR_PARSE or JB_ERR_MEMORY
2383 *********************************************************************/
2384 static jb_err receive_client_request(struct client_state *csp)
2386 char buf[BUFFER_SIZE];
2389 struct http_request *http;
2393 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2394 struct list header_list;
2395 struct list *headers = &header_list;
2399 memset(buf, 0, sizeof(buf));
2401 req = get_request_line(csp);
2404 return JB_ERR_PARSE;
2406 assert(*req != '\0');
2408 if (client_protocol_is_unsupported(csp, req))
2410 return JB_ERR_PARSE;
2413 #ifdef FEATURE_FORCE_LOAD
2415 * If this request contains the FORCE_PREFIX and blocks
2416 * aren't enforced, get rid of it and set the force flag.
2418 if (strstr(req, FORCE_PREFIX))
2420 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2422 log_error(LOG_LEVEL_FORCE,
2423 "Ignored force prefix in request: \"%s\".", req);
2427 strclean(req, FORCE_PREFIX);
2428 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2429 csp->flags |= CSP_FLAG_FORCED;
2432 #endif /* def FEATURE_FORCE_LOAD */
2434 err = parse_http_request(req, http);
2436 if (JB_ERR_OK != err)
2438 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2439 /* XXX: Use correct size */
2440 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2441 log_error(LOG_LEVEL_ERROR,
2442 "Couldn't parse request line received from %s: %s",
2443 csp->ip_addr_str, jb_err_to_string(err));
2445 free_http_request(http);
2446 return JB_ERR_PARSE;
2449 /* grab the rest of the client's headers */
2453 p = get_header(csp->iob);
2457 /* There are no additional headers to read. */
2464 * We didn't receive a complete header
2465 * line yet, get the rest of it.
2467 if (!data_is_available(csp->cfd, csp->config->socket_timeout))
2469 log_error(LOG_LEVEL_ERROR,
2470 "Stopped grabbing the client headers.");
2471 return JB_ERR_PARSE;
2474 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2477 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2478 destroy_list(headers);
2479 return JB_ERR_PARSE;
2482 if (add_to_iob(csp, buf, len))
2485 * If there is no memory left for buffering the
2486 * request, there is nothing we can do but hang up
2488 destroy_list(headers);
2489 return JB_ERR_MEMORY;
2495 * We were able to read a complete
2496 * header and can finaly enlist it.
2503 if (http->host == NULL)
2506 * If we still don't know the request destination,
2507 * the request is invalid or the client uses
2508 * Privoxy without its knowledge.
2510 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2513 * Our attempts to get the request destination
2514 * elsewhere failed or Privoxy is configured
2515 * to only accept proxy requests.
2517 * An error response has already been send
2518 * and we're done here.
2520 return JB_ERR_PARSE;
2525 * Determine the actions for this URL
2527 #ifdef FEATURE_TOGGLE
2528 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2530 /* Most compatible set of actions (i.e. none) */
2531 init_current_action(csp->action);
2534 #endif /* ndef FEATURE_TOGGLE */
2536 get_url_actions(csp, http);
2540 * Save a copy of the original request for logging
2542 http->ocmd = strdup(http->cmd);
2543 if (http->ocmd == NULL)
2545 log_error(LOG_LEVEL_FATAL,
2546 "Out of memory copying HTTP request line");
2548 enlist(csp->headers, http->cmd);
2550 /* Append the previously read headers */
2551 list_append_list_unique(csp->headers, headers);
2552 destroy_list(headers);
2559 /*********************************************************************
2561 * Function : parse_client_request
2563 * Description : Parses the client's request and decides what to do
2566 * Note that since we're not using select() we could get
2567 * blocked here if a client connected, then didn't say
2571 * 1 : csp = Current client state (buffers, headers, etc...)
2573 * Returns : JB_ERR_OK or JB_ERR_PARSE
2575 *********************************************************************/
2576 static jb_err parse_client_request(struct client_state *csp)
2578 struct http_request *http = csp->http;
2581 err = sed(csp, FILTER_CLIENT_HEADERS);
2582 if (JB_ERR_OK != err)
2584 /* XXX: Should be handled in sed(). */
2585 assert(err == JB_ERR_PARSE);
2586 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers.");
2588 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2590 /* Check request line for rewrites. */
2591 if ((NULL == csp->headers->first->str)
2592 || (strcmp(http->cmd, csp->headers->first->str) &&
2593 (JB_ERR_OK != change_request_destination(csp))))
2596 * A header filter broke the request line - bail out.
2598 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2599 /* XXX: Use correct size */
2600 log_error(LOG_LEVEL_CLF,
2601 "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2602 log_error(LOG_LEVEL_ERROR,
2603 "Invalid request line after applying header filters.");
2604 free_http_request(http);
2606 return JB_ERR_PARSE;
2614 /*********************************************************************
2618 * Description : Once a connection to the client has been accepted,
2619 * this function is called (via serve()) to handle the
2620 * main business of the communication. When this
2621 * function returns, the caller must close the client
2624 * FIXME: chat is nearly thousand lines long.
2628 * 1 : csp = Current client state (buffers, headers, etc...)
2630 * Returns : Nothing.
2632 *********************************************************************/
2633 static void chat(struct client_state *csp)
2635 char buf[BUFFER_SIZE];
2642 int ms_iis5_hack = 0;
2643 unsigned long long byte_count = 0;
2644 int forwarded_connect_retries = 0;
2645 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2646 const struct forward_spec *fwd;
2647 struct http_request *http;
2648 long len = 0; /* for buffer sizes (and negative error codes) */
2650 /* Function that does the content filtering for the current request */
2651 filter_function_ptr content_filter = NULL;
2653 /* Skeleton for HTTP response, if we should intercept the request */
2654 struct http_response *rsp;
2655 struct timeval timeout;
2657 memset(buf, 0, sizeof(buf));
2661 if (receive_client_request(csp) != JB_ERR_OK)
2665 if (parse_client_request(csp) != JB_ERR_OK)
2670 /* decide how to route the HTTP request */
2671 fwd = forward_url(csp, http);
2674 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2675 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2680 * build the http request to send to the server
2681 * we have to do one of the following:
2683 * create = use the original HTTP request to create a new
2684 * HTTP request that has either the path component
2685 * without the http://domainspec (w/path) or the
2686 * full orininal URL (w/url)
2687 * Note that the path and/or the HTTP version may
2688 * have been altered by now.
2690 * connect = Open a socket to the host:port of the server
2691 * and short-circuit server and client socket.
2693 * pass = Pass the request unchanged if forwarding a CONNECT
2694 * request to a parent proxy. Note that we'll be sending
2695 * the CFAIL message ourselves if connecting to the parent
2696 * fails, but we won't send a CSUCCEED message if it works,
2697 * since that would result in a double message (ours and the
2698 * parent's). After sending the request to the parent, we simply
2701 * here's the matrix:
2704 * +--------+--------+
2706 * 0 | create | connect|
2708 * Forwarding +--------+--------+
2710 * 1 | create | pass |
2712 * +--------+--------+
2716 if (http->ssl && connect_port_is_forbidden(csp))
2718 const char *acceptable_connect_ports =
2719 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2720 assert(NULL != acceptable_connect_ports);
2721 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2722 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2723 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2724 csp->action->flags |= ACTION_BLOCK;
2730 freez(csp->headers->first->str);
2731 build_request_line(csp, fwd, &csp->headers->first->str);
2735 * We have a request. Check if one of the crunchers wants it.
2737 if (crunch_response_triggered(csp, crunchers_all))
2740 * Yes. The client got the crunch response
2741 * and we are done here after cleaning up.
2743 /* XXX: why list_remove_all()? */
2744 list_remove_all(csp->headers);
2749 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2751 if (fwd->forward_host)
2753 log_error(LOG_LEVEL_CONNECT, "via [%s]:%d to: %s",
2754 fwd->forward_host, fwd->forward_port, http->hostport);
2758 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2761 /* here we connect to the server, gateway, or the forwarder */
2763 while ((csp->sfd = forwarded_connect(fwd, http, csp))
2764 && (errno == EINVAL)
2765 && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2767 log_error(LOG_LEVEL_ERROR,
2768 "failed request #%u to connect to %s. Trying again.",
2769 forwarded_connect_retries, http->hostport);
2772 if (csp->sfd == JB_INVALID_SOCKET)
2774 if (fwd->type != SOCKS_NONE)
2777 rsp = error_response(csp, "forwarding-failed", errno);
2779 else if (errno == EINVAL)
2781 rsp = error_response(csp, "no-such-domain", errno);
2785 rsp = error_response(csp, "connect-failed", errno);
2786 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2790 /* Write the answer to the client */
2793 send_crunch_response(csp, rsp);
2799 hdr = list_to_text(csp->headers);
2802 /* FIXME Should handle error properly */
2803 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2805 list_remove_all(csp->headers);
2807 if (fwd->forward_host || (http->ssl == 0))
2810 * Write the client's (modified) header to the server
2811 * (along with anything else that may be in the buffer)
2813 if (write_socket(csp->sfd, hdr, strlen(hdr))
2814 || (flush_socket(csp->sfd, csp->iob) < 0))
2816 log_error(LOG_LEVEL_CONNECT,
2817 "write header to: %s failed: %E", http->hostport);
2819 rsp = error_response(csp, "connect-failed", errno);
2822 send_crunch_response(csp, rsp);
2832 * We're running an SSL tunnel and we're not forwarding,
2833 * so just send the "connect succeeded" message to the
2834 * client, flush the rest, and get out of the way.
2836 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2844 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2846 /* we're finished with the client's header */
2849 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2851 /* pass data between the client and server
2852 * until one or the other shuts down the connection.
2861 * FD_ZERO here seems to point to an errant macro which crashes.
2862 * So do this by hand for now...
2864 memset(&rfds,0x00,sizeof(fd_set));
2868 FD_SET(csp->cfd, &rfds);
2869 FD_SET(csp->sfd, &rfds);
2871 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2872 if ((csp->flags & CSP_FLAG_CHUNKED)
2873 && !(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
2874 && ((csp->iob->eod - csp->iob->cur) >= 5)
2875 && !memcmp(csp->iob->eod-5, "0\r\n\r\n", 5))
2877 log_error(LOG_LEVEL_CONNECT,
2878 "Looks like we read the last chunk together with "
2879 "the server headers. We better stop reading.");
2880 byte_count = (unsigned long long)(csp->iob->eod - csp->iob->cur);
2881 csp->expected_content_length = byte_count;
2882 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2884 if (server_body && server_response_is_complete(csp, byte_count))
2886 log_error(LOG_LEVEL_CONNECT,
2887 "Done reading from server. Expected content length: %llu. "
2888 "Actual content length: %llu. Most recently received: %d.",
2889 csp->expected_content_length, byte_count, len);
2892 * XXX: should not jump around,
2893 * chat() is complicated enough already.
2897 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2899 timeout.tv_sec = csp->config->socket_timeout;
2900 timeout.tv_usec = 0;
2901 n = select((int)maxfd+1, &rfds, NULL, NULL, &timeout);
2905 log_error(LOG_LEVEL_ERROR,
2906 "Didn't receive data in time: %s", http->url);
2907 if ((byte_count == 0) && (http->ssl == 0))
2909 write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE,
2910 strlen(CONNECTION_TIMEOUT_RESPONSE));
2912 mark_server_socket_tainted(csp);
2917 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2918 mark_server_socket_tainted(csp);
2923 * This is the body of the browser's request,
2924 * just read and write it.
2926 if (FD_ISSET(csp->cfd, &rfds))
2928 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2932 /* XXX: not sure if this is necessary. */
2933 mark_server_socket_tainted(csp);
2934 break; /* "game over, man" */
2937 if (write_socket(csp->sfd, buf, (size_t)len))
2939 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2940 mark_server_socket_tainted(csp);
2947 * The server wants to talk. It could be the header or the body.
2948 * If `hdr' is null, then it's the header otherwise it's the body.
2949 * FIXME: Does `hdr' really mean `host'? No.
2951 if (FD_ISSET(csp->sfd, &rfds))
2954 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2958 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2960 if (http->ssl && (fwd->forward_host == NULL))
2963 * Just hang up. We already confirmed the client's CONNECT
2964 * request with status code 200 and unencrypted content is
2965 * no longer welcome.
2967 log_error(LOG_LEVEL_ERROR,
2968 "CONNECT already confirmed. Unable to tell the client about the problem.");
2971 else if (byte_count)
2974 * Just hang up. We already transmitted the original headers
2975 * and parts of the original content and therefore missed the
2976 * chance to send an error message (without risking data corruption).
2978 * XXX: we could retry with a fancy range request here.
2980 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2981 "Unable to tell the client about the problem.");
2982 mark_server_socket_tainted(csp);
2986 * XXX: Consider handling the cases above the same.
2988 mark_server_socket_tainted(csp);
2992 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2993 if (csp->flags & CSP_FLAG_CHUNKED)
2995 if ((len >= 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
2997 /* XXX: this is a temporary hack */
2998 log_error(LOG_LEVEL_CONNECT,
2999 "Looks like we reached the end of the last chunk. "
3000 "We better stop reading.");
3001 csp->expected_content_length = byte_count + (unsigned long long)len;
3002 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
3006 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
3009 * Add a trailing zero to let be able to use string operations.
3010 * XXX: do we still need this with filter_popups gone?
3015 * Normally, this would indicate that we've read
3016 * as much as the server has sent us and we can
3017 * close the client connection. However, Microsoft
3018 * in its wisdom has released IIS/5 with a bug that
3019 * prevents it from sending the trailing \r\n in
3020 * a 302 redirect header (and possibly other headers).
3021 * To work around this if we've haven't parsed
3022 * a full header we'll append a trailing \r\n
3023 * and see if this now generates a valid one.
3025 * This hack shouldn't have any impacts. If we've
3026 * already transmitted the header or if this is a
3027 * SSL connection, then we won't bother with this
3028 * hack. So we only work on partially received
3029 * headers. If we append a \r\n and this still
3030 * doesn't generate a valid header, then we won't
3031 * transmit anything to the client.
3036 if (server_body || http->ssl)
3039 * If we have been buffering up the document,
3040 * now is the time to apply content modification
3041 * and send the result to the client.
3045 p = execute_content_filter(csp, content_filter);
3047 * If the content filter fails, use the original
3048 * buffer and length.
3049 * (see p != NULL ? p : csp->iob->cur below)
3053 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
3056 if (JB_ERR_OK != update_server_headers(csp))
3058 log_error(LOG_LEVEL_FATAL,
3059 "Failed to update server headers. after filtering.");
3062 hdr = list_to_text(csp->headers);
3065 /* FIXME Should handle error properly */
3066 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3069 if (write_socket(csp->cfd, hdr, strlen(hdr))
3070 || write_socket(csp->cfd,
3071 ((p != NULL) ? p : csp->iob->cur), (size_t)csp->content_length))
3073 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
3076 mark_server_socket_tainted(csp);
3084 break; /* "game over, man" */
3088 * This is NOT the body, so
3089 * Let's pretend the server just sent us a blank line.
3091 snprintf(buf, sizeof(buf), "\r\n");
3092 len = (int)strlen(buf);
3095 * Now, let the normal header parsing algorithm below do its
3096 * job. If it fails, we'll exit instead of continuing.
3103 * If this is an SSL connection or we're in the body
3104 * of the server document, just write it to the client,
3105 * unless we need to buffer the body for later content-filtering
3107 if (server_body || http->ssl)
3112 * If there is no memory left for buffering the content, or the buffer limit
3113 * has been reached, switch to non-filtering mode, i.e. make & write the
3114 * header, flush the iob and buf, and get out of the way.
3116 if (add_to_iob(csp, buf, len))
3121 log_error(LOG_LEVEL_INFO,
3122 "Flushing header and buffers. Stepping back from filtering.");
3124 hdr = list_to_text(csp->headers);
3128 * Memory is too tight to even generate the header.
3129 * Send our static "Out-of-memory" page.
3131 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
3132 rsp = cgi_error_memory();
3133 send_crunch_response(csp, rsp);
3134 mark_server_socket_tainted(csp);
3137 hdrlen = strlen(hdr);
3139 if (write_socket(csp->cfd, hdr, hdrlen)
3140 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
3141 || (write_socket(csp->cfd, buf, (size_t)len)))
3143 log_error(LOG_LEVEL_CONNECT,
3144 "Flush header and buffers to client failed: %E");
3146 mark_server_socket_tainted(csp);
3151 * Reset the byte_count to the amount of bytes
3152 * we just flushed. len will be added a few lines below,
3153 * hdrlen doesn't matter for LOG_LEVEL_CLF.
3155 byte_count = (unsigned long long)flushed;
3157 content_filter = NULL;
3163 if (write_socket(csp->cfd, buf, (size_t)len))
3165 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
3166 mark_server_socket_tainted(csp);
3170 byte_count += (unsigned long long)len;
3175 const char *header_start;
3177 * We're still looking for the end of the server's header.
3178 * Buffer up the data we just read. If that fails, there's
3179 * little we can do but send our static out-of-memory page.
3181 if (add_to_iob(csp, buf, len))
3183 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
3184 rsp = cgi_error_memory();
3185 send_crunch_response(csp, rsp);
3186 mark_server_socket_tainted(csp);
3190 header_start = csp->iob->cur;
3192 /* Convert iob into something sed() can digest */
3193 if (JB_ERR_PARSE == get_server_headers(csp))
3198 * Well, we tried our MS IIS/5 hack and it didn't work.
3199 * The header is incomplete and there isn't anything
3200 * we can do about it.
3202 log_error(LOG_LEVEL_ERROR, "Invalid server headers. "
3203 "Applying the MS IIS5 hack didn't help.");
3204 log_error(LOG_LEVEL_CLF,
3205 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3206 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3207 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3208 mark_server_socket_tainted(csp);
3214 * Since we have to wait for more from the server before
3215 * we can parse the headers we just continue here.
3217 long header_offset = csp->iob->cur - header_start;
3218 assert(csp->iob->cur >= header_start);
3219 byte_count += (unsigned long long)(len - header_offset);
3220 log_error(LOG_LEVEL_CONNECT, "Continuing buffering headers. "
3221 "byte_count: %llu. header_offset: %d. len: %d.",
3222 byte_count, header_offset, len);
3227 /* Did we actually get anything? */
3228 if (NULL == csp->headers->first)
3230 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
3231 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3232 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
3233 free_http_request(http);
3234 mark_server_socket_tainted(csp);
3238 assert(csp->headers->first->str);
3240 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
3241 strncmpic(csp->headers->first->str, "ICY", 3))
3244 * It doesn't look like a HTTP (or Shoutcast) response:
3245 * tell the client and log the problem.
3247 if (strlen(csp->headers->first->str) > 30)
3249 csp->headers->first->str[30] = '\0';
3251 log_error(LOG_LEVEL_ERROR,
3252 "Invalid server or forwarder response. Starts with: %s",
3253 csp->headers->first->str);
3254 log_error(LOG_LEVEL_CLF,
3255 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3256 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3257 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3258 free_http_request(http);
3259 mark_server_socket_tainted(csp);
3264 * We have now received the entire server header,
3265 * filter it and send the result to the client
3267 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
3269 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
3271 hdr = list_to_text(csp->headers);
3274 /* FIXME Should handle error properly */
3275 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3278 if (crunch_response_triggered(csp, crunchers_light))
3281 * One of the tags created by a server-header
3282 * tagger triggered a crunch. We already
3283 * delivered the crunch response to the client
3284 * and are done here after cleaning up.
3287 mark_server_socket_tainted(csp);
3290 /* Buffer and pcrs filter this if appropriate. */
3292 if (!http->ssl) /* We talk plaintext */
3294 content_filter = get_filter_function(csp);
3297 * Only write if we're not buffering for content modification
3299 if (!content_filter)
3302 * Write the server's (modified) header to
3303 * the client (along with anything else that
3304 * may be in the buffer)
3307 if (write_socket(csp->cfd, hdr, strlen(hdr))
3308 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
3310 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
3313 * The write failed, so don't bother mentioning it
3314 * to the client... it probably can't hear us anyway.
3317 mark_server_socket_tainted(csp);
3321 byte_count += (unsigned long long)len;
3326 * XXX: the header lenght should probably
3327 * be calculated by get_server_headers().
3329 long header_length = csp->iob->cur - header_start;
3330 assert(csp->iob->cur > header_start);
3331 byte_count += (unsigned long long)(len - header_length);
3334 /* we're finished with the server's header */
3340 * If this was a MS IIS/5 hack then it means the server
3341 * has already closed the connection. Nothing more to read.
3346 log_error(LOG_LEVEL_ERROR,
3347 "Closed server connection detected. "
3348 "Applying the MS IIS5 hack didn't help.");
3349 log_error(LOG_LEVEL_CLF,
3350 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3351 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3352 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3353 mark_server_socket_tainted(csp);
3359 mark_server_socket_tainted(csp);
3360 return; /* huh? we should never get here */
3363 if (csp->content_length == 0)
3366 * If Privoxy didn't recalculate the Content-Lenght,
3367 * byte_count is still correct.
3369 csp->content_length = byte_count;
3372 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3373 if ((csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
3374 && (csp->expected_content_length != byte_count))
3376 log_error(LOG_LEVEL_CONNECT,
3377 "Received %llu bytes while expecting %llu.",
3378 byte_count, csp->expected_content_length);
3379 mark_server_socket_tainted(csp);
3383 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %llu",
3384 csp->ip_addr_str, http->ocmd, csp->content_length);
3388 /*********************************************************************
3392 * Description : This is little more than chat. We only "serve" to
3393 * to close (or remember) any socket that chat may have
3397 * 1 : csp = Current client state (buffers, headers, etc...)
3401 *********************************************************************/
3403 void serve(struct client_state *csp)
3404 #else /* ifndef AMIGA */
3405 static void serve(struct client_state *csp)
3406 #endif /* def AMIGA */
3410 if (csp->sfd != JB_INVALID_SOCKET)
3412 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3413 static int monitor_thread_running = 0;
3415 if ((csp->config->feature_flags & RUNTIME_FEATURE_CONNECTION_KEEP_ALIVE)
3416 && (csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
3418 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
3419 close_socket(csp->cfd);
3420 csp->cfd = JB_INVALID_SOCKET;
3421 privoxy_mutex_lock(&connection_reuse_mutex);
3422 if (!monitor_thread_running)
3424 monitor_thread_running = 1;
3425 privoxy_mutex_unlock(&connection_reuse_mutex);
3426 wait_for_alive_connections();
3427 privoxy_mutex_lock(&connection_reuse_mutex);
3428 monitor_thread_running = 0;
3430 privoxy_mutex_unlock(&connection_reuse_mutex);
3434 forget_connection(csp->sfd);
3435 close_socket(csp->sfd);
3438 close_socket(csp->sfd);
3439 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3442 if (csp->cfd != JB_INVALID_SOCKET)
3444 close_socket(csp->cfd);
3447 csp->flags &= ~CSP_FLAG_ACTIVE;
3453 /*********************************************************************
3455 * Function : server_thread
3457 * Description : We only exist to call `serve' in a threaded environment.
3460 * 1 : data = Current client state (buffers, headers, etc...)
3462 * Returns : Always 0.
3464 *********************************************************************/
3465 static int32 server_thread(void *data)
3467 serve((struct client_state *) data);
3474 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3475 /*********************************************************************
3479 * Description : Print usage info & exit.
3481 * Parameters : Pointer to argv[0] for identifying ourselves
3485 *********************************************************************/
3486 static void usage(const char *myname)
3488 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3492 #endif /* defined(unix) */
3495 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3496 #endif /* defined(unix) */
3497 "[--version] [configfile]\n"
3498 "Aborting\n", myname);
3503 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3506 #ifdef MUTEX_LOCKS_AVAILABLE
3507 /*********************************************************************
3509 * Function : privoxy_mutex_lock
3511 * Description : Locks a mutex.
3514 * 1 : mutex = The mutex to lock.
3516 * Returns : Void. May exit in case of errors.
3518 *********************************************************************/
3519 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3521 #ifdef FEATURE_PTHREAD
3522 int err = pthread_mutex_lock(mutex);
3525 if (mutex != &log_mutex)
3527 log_error(LOG_LEVEL_FATAL,
3528 "Mutex locking failed: %s.\n", strerror(err));
3533 EnterCriticalSection(mutex);
3534 #endif /* def FEATURE_PTHREAD */
3538 /*********************************************************************
3540 * Function : privoxy_mutex_unlock
3542 * Description : Unlocks a mutex.
3545 * 1 : mutex = The mutex to unlock.
3547 * Returns : Void. May exit in case of errors.
3549 *********************************************************************/
3550 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3552 #ifdef FEATURE_PTHREAD
3553 int err = pthread_mutex_unlock(mutex);
3556 if (mutex != &log_mutex)
3558 log_error(LOG_LEVEL_FATAL,
3559 "Mutex unlocking failed: %s.\n", strerror(err));
3564 LeaveCriticalSection(mutex);
3565 #endif /* def FEATURE_PTHREAD */
3569 /*********************************************************************
3571 * Function : privoxy_mutex_init
3573 * Description : Prepares a mutex.
3576 * 1 : mutex = The mutex to initialize.
3578 * Returns : Void. May exit in case of errors.
3580 *********************************************************************/
3581 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3583 #ifdef FEATURE_PTHREAD
3584 int err = pthread_mutex_init(mutex, 0);
3587 printf("Fatal error. Mutex initialization failed: %s.\n",
3592 InitializeCriticalSection(mutex);
3593 #endif /* def FEATURE_PTHREAD */
3595 #endif /* def MUTEX_LOCKS_AVAILABLE */
3597 /*********************************************************************
3599 * Function : initialize_mutexes
3601 * Description : Prepares mutexes if mutex support is available.
3605 * Returns : Void, exits in case of errors.
3607 *********************************************************************/
3608 static void initialize_mutexes(void)
3610 #ifdef MUTEX_LOCKS_AVAILABLE
3612 * Prepare global mutex semaphores
3614 privoxy_mutex_init(&log_mutex);
3615 privoxy_mutex_init(&log_init_mutex);
3616 privoxy_mutex_init(&connection_reuse_mutex);
3619 * XXX: The assumptions below are a bit naive
3620 * and can cause locks that aren't necessary.
3622 * For example older FreeBSD versions (< 6.x?)
3623 * have no gethostbyname_r, but gethostbyname is
3626 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3627 privoxy_mutex_init(&resolver_mutex);
3628 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3630 * XXX: should we use a single mutex for
3631 * localtime() and gmtime() as well?
3633 #ifndef HAVE_GMTIME_R
3634 privoxy_mutex_init(&gmtime_mutex);
3635 #endif /* ndef HAVE_GMTIME_R */
3637 #ifndef HAVE_LOCALTIME_R
3638 privoxy_mutex_init(&localtime_mutex);
3639 #endif /* ndef HAVE_GMTIME_R */
3642 privoxy_mutex_init(&rand_mutex);
3643 #endif /* ndef HAVE_RANDOM */
3644 #endif /* def MUTEX_LOCKS_AVAILABLE */
3648 /*********************************************************************
3652 * Description : Load the config file and start the listen loop.
3653 * This function is a lot more *sane* with the `load_config'
3654 * and `listen_loop' functions; although it stills does
3655 * a *little* too much for my taste.
3658 * 1 : argc = Number of parameters (including $0).
3659 * 2 : argv = Array of (char *)'s to the parameters.
3661 * Returns : 1 if : can't open config file, unrecognized directive,
3662 * stats requested in multi-thread mode, can't open the
3663 * log file, can't open the jar file, listen port is invalid,
3664 * any load fails, and can't bind port.
3666 * Else main never returns, the process must be signaled
3667 * to terminate execution. Or, on Windows, use the
3668 * "File", "Exit" menu option.
3670 *********************************************************************/
3672 int real_main(int argc, const char *argv[])
3674 int main(int argc, const char *argv[])
3678 unsigned int random_seed;
3680 struct passwd *pw = NULL;
3681 struct group *grp = NULL;
3684 char *pre_chroot_nslookup_to_load_resolver = NULL;
3691 #if !defined(_WIN32)
3698 /* Prepare mutexes if supported and necessary. */
3699 initialize_mutexes();
3701 /* Enable logging until further notice. */
3705 * Parse the command line arguments
3707 * XXX: simply printing usage information in case of
3708 * invalid arguments isn't particularly user friendly.
3710 while (++argc_pos < argc)
3713 /* Check to see if the service must be installed or uninstalled */
3714 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3716 const char *pName = argv[argc_pos] + 9;
3719 exit( (install_service(pName)) ? 0 : 1 );
3721 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3723 const char *pName = argv[argc_pos] + 11;
3726 exit((uninstall_service(pName)) ? 0 : 1);
3728 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3730 bRunAsService = TRUE;
3731 w32_set_service_cwd();
3732 atexit(w32_service_exit_notify);
3735 #endif /* defined(_WIN32) */
3738 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3740 if (strcmp(argv[argc_pos], "--help") == 0)
3745 else if(strcmp(argv[argc_pos], "--version") == 0)
3747 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3753 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3755 set_debug_level(LOG_LEVEL_FATAL | LOG_LEVEL_ERROR | LOG_LEVEL_INFO);
3759 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3761 if (++argc_pos == argc) usage(argv[0]);
3762 pidfile = strdup(argv[argc_pos]);
3765 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3767 if (++argc_pos == argc) usage(argv[argc_pos]);
3769 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3772 if (NULL == (grp = getgrnam(p)))
3774 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3778 if (NULL == (pw = getpwnam(argv[argc_pos])))
3780 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3783 if (p != NULL) *--p = '\0';
3786 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3788 if (++argc_pos == argc) usage(argv[0]);
3789 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3792 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3796 #endif /* defined(unix) */
3798 else if (argc_pos + 1 != argc)
3801 * This is neither the last command line
3802 * option, nor was it recognized before,
3803 * therefore it must be invalid.
3809 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3811 configfile = argv[argc_pos];
3814 } /* -END- while (more arguments) */
3816 show_version(Argv[0]);
3819 if ( *configfile != '/' )
3821 char cwd[BUFFER_SIZE];
3823 size_t abs_file_size;
3825 /* make config-filename absolute here */
3826 if (NULL == getcwd(cwd, sizeof(cwd)))
3828 perror("failed to get current working directory");
3833 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3834 basedir = strdup(cwd);
3836 if (NULL == basedir ||
3837 NULL == (abs_file = malloc(abs_file_size)))
3839 perror("malloc failed");
3842 strlcpy(abs_file, basedir, abs_file_size);
3843 strlcat(abs_file, "/", abs_file_size );
3844 strlcat(abs_file, configfile, abs_file_size);
3845 configfile = abs_file;
3847 #endif /* defined unix */
3851 clients->next = NULL;
3853 /* XXX: factor out initialising after the next stable release. */
3856 #elif defined(_WIN32)
3860 random_seed = (unsigned int)time(NULL);
3862 srandom(random_seed);
3865 #endif /* ifdef HAVE_RANDOM */
3868 * Unix signal handling
3870 * Catch the abort, interrupt and terminate signals for a graceful exit
3871 * Catch the hangup signal so the errlog can be reopened.
3872 * Ignore the broken pipe signals (FIXME: Why?)
3874 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3877 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3878 const int ignored_signals[] = { SIGPIPE, 0 };
3880 for (idx = 0; catched_signals[idx] != 0; idx++)
3882 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3883 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3885 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3886 #endif /* ifdef sun */
3888 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3892 for (idx = 0; ignored_signals[idx] != 0; idx++)
3894 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3896 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3901 #else /* ifdef _WIN32 */
3902 # ifdef _WIN_CONSOLE
3904 * We *are* in a windows console app.
3905 * Print a verbose messages about FAQ's and such
3907 printf("%s", win32_blurb);
3908 # endif /* def _WIN_CONSOLE */
3909 #endif /* def _WIN32 */
3912 /* Initialize the CGI subsystem */
3913 cgi_init_error_messages();
3916 * If runnig on unix and without the --nodaemon
3917 * option, become a daemon. I.e. fork, detach
3918 * from tty and get process group leadership
3931 if ( pid < 0 ) /* error */
3936 else if ( pid != 0 ) /* parent */
3941 * must check for errors
3942 * child died due to missing files aso
3945 wpid = waitpid( pid, &status, WNOHANG );
3954 /* Should be more portable, but not as well tested */
3959 #else /* ndef __FreeBSD__ */
3961 #endif /* ndef __FreeBSD__ */
3962 fd = open("/dev/tty", O_RDONLY);
3965 /* no error check here */
3966 ioctl( fd, TIOCNOTTY,0 );
3971 * stderr (fd 2) will be closed later on,
3972 * when the config file has been parsed.
3979 } /* -END- if (!no_daemon) */
3982 * As soon as we have written the PID file, we can switch
3983 * to the user and group ID indicated by the --user option
3989 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3991 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3995 if (setgroups(1, &grp->gr_gid))
3997 log_error(LOG_LEVEL_FATAL, "setgroups() failed: %E");
4000 else if (initgroups(pw->pw_name, pw->pw_gid))
4002 log_error(LOG_LEVEL_FATAL, "initgroups() failed: %E");
4008 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
4010 /* Read the time zone file from /etc before doing chroot. */
4012 if (NULL != pre_chroot_nslookup_to_load_resolver
4013 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
4015 /* Initialize resolver library. */
4016 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
4018 if (chroot(pw->pw_dir) < 0)
4020 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
4024 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
4027 if (setuid(pw->pw_uid))
4029 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
4033 char putenv_dummy[64];
4035 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
4036 if (putenv(putenv_dummy) != 0)
4038 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
4041 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
4042 if (putenv(putenv_dummy) != 0)
4044 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
4050 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
4053 #endif /* defined unix */
4056 /* This will be FALSE unless the command line specified --service
4060 /* Yup, so now we must attempt to establish a connection
4061 * with the service dispatcher. This will only work if this
4062 * process was launched by the service control manager to
4063 * actually run as a service. If this isn't the case, i've
4064 * known it take around 30 seconds or so for the call to return.
4067 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
4068 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
4070 /* Service has run, and at this point is now being stopped, so just return */
4075 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
4077 /* An error occurred. Usually it's because --service was wrongly specified
4078 * and we were unable to connect to the Service Control Dispatcher because
4079 * it wasn't expecting us and is therefore not listening.
4081 * For now, just continue below to call the listen_loop function.
4084 #endif /* def _WIN32 */
4094 /*********************************************************************
4096 * Function : bind_port_helper
4098 * Description : Bind the listen port. Handles logging, and aborts
4102 * 1 : config = Privoxy configuration. Specifies port
4105 * Returns : Port that was opened.
4107 *********************************************************************/
4108 static jb_socket bind_port_helper(struct configuration_spec * config)
4113 if (config->haddr == NULL)
4115 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
4120 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
4121 config->hport, config->haddr);
4124 result = bind_port(config->haddr, config->hport, &bfd);
4131 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
4132 "There may be another Privoxy or some other "
4133 "proxy running on port %d",
4134 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
4135 config->hport, config->hport);
4138 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
4139 "The hostname is not resolvable",
4140 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
4143 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
4144 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
4147 /* shouldn't get here */
4148 return JB_INVALID_SOCKET;
4151 config->need_bind = 0;
4158 /* Without this simple workaround we get this compiler warning from _beginthread
4159 * warning C4028: formal parameter 1 different from declaration
4161 void w32_service_listen_loop(void *p)
4165 #endif /* def _WIN32 */
4168 /*********************************************************************
4170 * Function : listen_loop
4172 * Description : bind the listen port and enter a "FOREVER" listening loop.
4178 *********************************************************************/
4179 static void listen_loop(void)
4181 struct client_state *csp = NULL;
4183 struct configuration_spec * config;
4185 config = load_config();
4187 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
4189 * XXX: Should be relocated once it no
4190 * longer needs to emit log messages.
4192 initialize_reusable_connections();
4193 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
4195 bfd = bind_port_helper(config);
4197 #ifdef FEATURE_GRACEFUL_TERMINATION
4198 while (!g_terminate)
4203 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
4204 while (waitpid(-1, NULL, WNOHANG) > 0)
4206 /* zombie children */
4208 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
4211 * Free data that was used by died threads
4217 * Re-open the errlog after HUP signal
4219 if (received_hup_signal)
4221 if (NULL != config->logfile)
4223 init_error_log(Argv[0], config->logfile);
4225 received_hup_signal = 0;
4229 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
4231 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
4235 csp->flags |= CSP_FLAG_ACTIVE;
4236 csp->sfd = JB_INVALID_SOCKET;
4238 csp->config = config = load_config();
4240 if ( config->need_bind )
4243 * Since we were listening to the "old port", we will not see
4244 * a "listen" param change until the next request. So, at
4245 * least 1 more request must be made for us to find the new
4246 * setting. I am simply closing the old socket and binding the
4249 * Which-ever is correct, we will serve 1 more page via the
4250 * old settings. This should probably be a "show-proxy-args"
4251 * request. This should not be a so common of an operation
4252 * that this will hurt people's feelings.
4257 bfd = bind_port_helper(config);
4260 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
4262 if (!accept_connection(csp, bfd))
4264 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
4277 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
4280 #ifdef FEATURE_TOGGLE
4281 if (global_toggle_state)
4282 #endif /* def FEATURE_TOGGLE */
4284 csp->flags |= CSP_FLAG_TOGGLED_ON;
4287 if (run_loader(csp))
4289 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
4290 /* Never get here - LOG_LEVEL_FATAL causes program exit */
4294 if (block_acl(NULL,csp))
4296 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
4297 close_socket(csp->cfd);
4298 freez(csp->ip_addr_str);
4302 #endif /* def FEATURE_ACL */
4304 /* add it to the list of clients */
4305 csp->next = clients->next;
4306 clients->next = csp;
4308 if (config->multi_threaded)
4312 /* this is a switch () statment in the C preprocessor - ugh */
4313 #undef SELECTED_ONE_OPTION
4315 /* Use Pthreads in preference to native code */
4316 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
4317 #define SELECTED_ONE_OPTION
4319 pthread_t the_thread;
4320 pthread_attr_t attrs;
4322 pthread_attr_init(&attrs);
4323 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
4324 errno = pthread_create(&the_thread, &attrs,
4325 (void * (*)(void *))serve, csp);
4326 child_id = errno ? -1 : 0;
4327 pthread_attr_destroy(&attrs);
4331 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
4332 #define SELECTED_ONE_OPTION
4333 child_id = _beginthread(
4334 (void (*)(void *))serve,
4339 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
4340 #define SELECTED_ONE_OPTION
4341 child_id = _beginthread(
4342 (void(* _Optlink)(void*))serve,
4348 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
4349 #define SELECTED_ONE_OPTION
4351 thread_id tid = spawn_thread
4352 (server_thread, "server", B_NORMAL_PRIORITY, csp);
4354 if ((tid >= 0) && (resume_thread(tid) == B_OK))
4356 child_id = (int) tid;
4365 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
4366 #define SELECTED_ONE_OPTION
4367 csp->cfd = ReleaseSocket(csp->cfd, -1);
4370 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
4371 NP_Output, Output(),
4372 NP_CloseOutput, FALSE,
4373 NP_Name, (ULONG)"privoxy child",
4377 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
4378 NP_Output, Output(),
4379 NP_CloseOutput, FALSE,
4380 NP_Name, (ULONG)"privoxy child",
4381 NP_StackSize, 200*1024,
4387 ((struct Task *)child_id)->tc_UserData = csp;
4388 Signal((struct Task *)child_id, SIGF_SINGLE);
4393 #if !defined(SELECTED_ONE_OPTION)
4396 /* This block is only needed when using fork().
4397 * When using threads, the server thread was
4398 * created and run by the call to _beginthread().
4400 if (child_id == 0) /* child */
4403 #ifdef FEATURE_TOGGLE
4404 int inherited_toggle_state = global_toggle_state;
4405 #endif /* def FEATURE_TOGGLE */
4410 * If we've been toggled or we've blocked the request, tell Mom
4413 #ifdef FEATURE_TOGGLE
4414 if (inherited_toggle_state != global_toggle_state)
4416 rc |= RC_FLAG_TOGGLED;
4418 #endif /* def FEATURE_TOGGLE */
4420 #ifdef FEATURE_STATISTICS
4421 if (csp->flags & CSP_FLAG_REJECTED)
4423 rc |= RC_FLAG_BLOCKED;
4425 #endif /* ndef FEATURE_STATISTICS */
4429 else if (child_id > 0) /* parent */
4431 /* in a fork()'d environment, the parent's
4432 * copy of the client socket and the CSP
4436 #if !defined(_WIN32) && !defined(__CYGWIN__)
4438 wait( &child_status );
4441 * Evaluate child's return code: If the child has
4442 * - been toggled, toggle ourselves
4443 * - blocked its request, bump up the stats counter
4446 #ifdef FEATURE_TOGGLE
4447 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
4449 global_toggle_state = !global_toggle_state;
4451 #endif /* def FEATURE_TOGGLE */
4453 #ifdef FEATURE_STATISTICS
4455 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4459 #endif /* def FEATURE_STATISTICS */
4461 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4462 close_socket(csp->cfd);
4463 csp->flags &= ~CSP_FLAG_ACTIVE;
4467 #undef SELECTED_ONE_OPTION
4468 /* end of cpp switch () */
4473 * Spawning the child failed, assume it's because
4474 * there are too many children running already.
4475 * XXX: If you assume ...
4477 log_error(LOG_LEVEL_ERROR,
4478 "Unable to take any additional connections: %E");
4479 write_socket(csp->cfd, TOO_MANY_CONNECTIONS_RESPONSE,
4480 strlen(TOO_MANY_CONNECTIONS_RESPONSE));
4481 close_socket(csp->cfd);
4482 csp->flags &= ~CSP_FLAG_ACTIVE;
4491 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4493 /* Clean up. Aim: free all memory (no leaks) */
4494 #ifdef FEATURE_GRACEFUL_TERMINATION
4496 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4498 unload_current_config_file();
4499 unload_current_actions_file();
4500 unload_current_re_filterfile();
4501 #ifdef FEATURE_TRUST
4502 unload_current_trust_file();
4505 if (config->multi_threaded)
4512 } while ((clients->next != NULL) && (--i > 0));
4516 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4527 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4528 /* Cleanup - remove taskbar icon etc. */
4533 #endif /* FEATURE_GRACEFUL_TERMINATION */