1 const char jcc_rcs[] = "$Id: jcc.c,v 1.245 2009/04/24 15:29:43 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 1.245 2009/04/24 15:29:43 fabiankeil
37 * Allow to limit the number of of client connections.
39 * Revision 1.244 2009/04/17 11:34:34 fabiankeil
40 * Style cosmetics for the IPv6 code.
42 * Revision 1.243 2009/04/17 11:27:49 fabiankeil
43 * Petr Pisar's privoxy-3.0.12-ipv6-3.diff.
45 * Revision 1.242 2009/04/11 10:44:47 fabiankeil
46 * Update a comment. We're not in Kansas anymore.
48 * Revision 1.241 2009/04/11 10:37:23 fabiankeil
49 * When dropping connections due to ACL, don't leak csp->ip_addr_str.
51 * Revision 1.240 2009/04/09 10:12:54 fabiankeil
52 * Fix two cases in which an invalid server response would result
53 * in the client connection being closed without sending an error
56 * Revision 1.239 2009/04/07 11:43:50 fabiankeil
57 * If the server rudely resets the connection directly after sending the
58 * headers, pass the mess to the client instead of sending an incorrect
59 * connect-failed message. Fixes #2698674 reported by mybugaccount.
61 * Revision 1.238 2009/03/27 14:42:30 fabiankeil
62 * Correct the status code for CONNECTION_TIMEOUT_RESPONSE.
64 * Revision 1.237 2009/03/27 14:32:04 fabiankeil
65 * If spawning a child in listen_loop() fails, send a real
66 * HTTP response to the client and continue listening for
67 * new connections without artificial delay.
69 * Revision 1.236 2009/03/25 17:30:24 fabiankeil
70 * In serve(), keep the client socket open until we marked the
71 * server socket as unused. This should increase the chances
72 * that we reuse the connection for the client's next request
73 * to the same destination.
75 * Revision 1.235 2009/03/18 21:01:20 fabiankeil
76 * Comment fix. Spotted by Roland.
78 * Revision 1.234 2009/03/18 20:48:42 fabiankeil
79 * If the --no-daemon option is used, enable LOG_LEVEL_INFO
80 * before the config file has been parsed (as we always did).
82 * Revision 1.233 2009/03/13 14:10:07 fabiankeil
83 * Fix some more harmless warnings on amd64.
85 * Revision 1.232 2009/03/08 19:29:16 fabiankeil
86 * Reinitialize the timeout structure every time before passing
87 * it to select(). Apparently some implementations mess with it.
88 * Probably fixes #2669131 reported by cyberpatrol.
90 * Revision 1.231 2009/03/08 14:19:23 fabiankeil
91 * Fix justified (but harmless) compiler warnings
92 * on platforms where sizeof(int) < sizeof(long).
94 * Revision 1.230 2009/03/07 13:09:17 fabiankeil
95 * Change csp->expected_content and_csp->expected_content_length from
96 * size_t to unsigned long long to reduce the likelihood of integer
97 * overflows that would let us close the connection prematurely.
98 * Bug found while investigating #2669131, reported by cyberpatrol.
100 * Revision 1.229 2009/03/07 11:17:01 fabiankeil
101 * Fix compiler warning.
103 * Revision 1.228 2009/03/06 20:30:13 fabiankeil
104 * Log unsigned values as such.
106 * Revision 1.227 2009/03/02 19:18:11 fabiankeil
107 * Streamline parse_http_request()'s prototype. As
108 * cparser pointed out it doesn't actually use csp.
110 * Revision 1.226 2009/03/01 18:28:24 fabiankeil
111 * Help clang understand that we aren't dereferencing
112 * NULL pointers here.
114 * Revision 1.225 2009/02/19 18:09:32 fabiankeil
115 * Unbreak build without FEATURE_CONNECTION_KEEP_ALIVE.
118 * Revision 1.224 2009/02/14 15:32:04 fabiankeil
119 * Add the request URL to the timeout message in chat().
122 * Revision 1.223 2009/02/09 21:21:16 fabiankeil
123 * Now that init_log_module() is called earlier, call show_version()
124 * later on from main() directly so it doesn't get called for --help
127 * Revision 1.222 2009/02/08 12:56:51 fabiankeil
128 * Call initialize_mutexes() before init_log_module() again.
129 * Broken since r220, might be the cause of Lee's #2579448.
131 * Revision 1.221 2009/02/06 18:02:58 fabiankeil
132 * When dropping privileges, also give up membership in supplementary
133 * groups. Thanks to Matthias Drochner for reporting the problem,
134 * providing the initial patch and testing the final version.
136 * Revision 1.220 2009/02/04 18:29:07 fabiankeil
137 * Initialize the log module before parsing arguments.
138 * Thanks to Matthias Drochner for the report.
140 * Revision 1.219 2009/01/31 16:08:21 fabiankeil
141 * Remove redundant error check in receive_client_request().
143 * Revision 1.218 2009/01/31 12:25:54 fabiankeil
144 * Flatten indentation in receive_client_request().
146 * Revision 1.217 2009/01/07 19:50:09 fabiankeil
147 * - If the socket-timeout has been reached and the client
148 * hasn't received any data yet, send an explanation before
149 * closing the connection.
150 * - In get_request_line(), signal timeouts the right way.
152 * Revision 1.216 2008/12/24 22:13:11 ler762
153 * fix GCC 3.4.4 warning
155 * Revision 1.215 2008/12/24 17:06:19 fabiankeil
156 * Keep a thread around to timeout alive connections
157 * even if no new requests are coming in.
159 * Revision 1.214 2008/12/20 14:53:55 fabiankeil
160 * Add config option socket-timeout to control the time
161 * Privoxy waits for data to arrive on a socket. Useful
162 * in case of stale ssh tunnels or when fuzz-testing.
164 * Revision 1.213 2008/12/15 18:45:51 fabiankeil
165 * When logging crunches, log the whole URL, so one can easily
166 * differentiate between vanilla HTTP and CONNECT requests.
168 * Revision 1.212 2008/12/14 15:46:22 fabiankeil
169 * Give crunched requests their own log level.
171 * Revision 1.211 2008/12/06 10:05:03 fabiankeil
172 * Downgrade "Received x bytes while expecting y." message to
173 * LOG_LEVEL_CONNECT as it doesn't necessarily indicate an error.
175 * Revision 1.210 2008/12/02 22:03:18 fabiankeil
176 * Don't miscalculate byte_count if we don't get all the
177 * server headers with one read_socket() call. With keep-alive
178 * support enabled, this caused delays until the server closed
181 * Revision 1.209 2008/11/27 09:44:04 fabiankeil
182 * Cosmetics for the last commit: Don't watch out for
183 * the last chunk if the content isn't chunk-encoded or
184 * if we already determined the content length previously.
186 * Revision 1.208 2008/11/26 18:24:17 fabiankeil
187 * Recognize that the server response is complete if the
188 * last chunk is read together with the server headers.
191 * Revision 1.207 2008/11/25 17:25:16 fabiankeil
192 * Don't convert the client-header list to text until we need to.
194 * Revision 1.206 2008/11/23 17:00:11 fabiankeil
195 * Some more chat() cosmetics.
197 * Revision 1.205 2008/11/16 12:43:49 fabiankeil
198 * Turn keep-alive support into a runtime feature
199 * that is disabled by setting keep-alive-timeout
200 * to a negative value.
202 * Revision 1.204 2008/11/06 19:42:17 fabiankeil
203 * Fix last-chunk detection hack to also apply
204 * if buf[] contains nothing but the last-chunk.
206 * Revision 1.203 2008/11/06 18:34:35 fabiankeil
207 * Factor receive_client_request() and
208 * parse_client_request() out of chat().
210 * Revision 1.202 2008/11/02 18:40:34 fabiankeil
211 * If we received a different amount of data than we expected,
212 * log a warning and make sure the server socket isn't reused.
214 * Revision 1.201 2008/11/02 16:48:20 fabiankeil
215 * Revert revision 1.195 and try again.
217 * Revision 1.200 2008/10/26 16:53:18 fabiankeil
220 * Revision 1.199 2008/10/26 15:36:10 fabiankeil
221 * Remove two debug messages with LOG_LEVEL_INFO.
223 * Revision 1.198 2008/10/22 15:19:55 fabiankeil
224 * Once More, With Feeling: if there is no logfile
225 * because the user didn't specify one, we shouldn't
226 * call init_error_log() after receiving SIGHUP either.
228 * Revision 1.197 2008/10/20 17:02:40 fabiankeil
229 * If SIGHUP is received while we aren't running in daemon
230 * mode, calling init_error_log() would be a mistake.
232 * Revision 1.196 2008/10/16 09:16:41 fabiankeil
233 * - Fix two gcc44 conversion warnings.
234 * - Don't bother logging the last five bytes
237 * Revision 1.195 2008/10/13 16:04:37 fabiankeil
238 * Make sure we don't try to reuse tainted server sockets.
240 * Revision 1.194 2008/10/12 18:35:18 fabiankeil
241 * The last commit was a bit too ambitious, apparently the content
242 * length adjustment is only necessary if we aren't buffering.
244 * Revision 1.193 2008/10/12 15:57:35 fabiankeil
245 * Fix content length calculation if we read headers
246 * and the start of the body at once. Now that we have
247 * FEATURE_CONNECTION_KEEP_ALIVE, it actually matters.
249 * Revision 1.192 2008/10/11 18:19:14 fabiankeil
250 * Even more chat() cosmetics.
252 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
253 * Reformat some comments in chat().
255 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
256 * In case of chunk-encoded content, stop reading if
257 * the buffer looks like it ends with the last chunk.
259 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
260 * Let server_response_is_complete() deal properly with
261 * content that is neither buffered nor read all at once.
263 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
264 * Flush work-in-progress changes to keep outgoing connections
265 * alive where possible. Incomplete and mostly #ifdef'd out.
267 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
268 * Add mutex lock support for _WIN32.
270 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
271 * Prepare for critical sections on Windows by adding a
272 * layer of indirection before the pthread mutex functions.
274 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
275 * Remove FEATURE_COOKIE_JAR.
277 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
278 * - Silence LLVM/Clang complaint.
279 * - Make received_hup_signal static.
280 * - Hide definitions for basedir, pidfile and received_hup_signal
281 * from __EMX__ as they only seem to be used in case of #ifdef unix.
283 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
284 * Accept Shoutcast responses again. Problem reported
285 * and fix suggested by Stefan in #2062860.
287 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
288 * Fix possible NULL-pointer dereference reported
289 * by din_a4 in #2003937. Pointy hat to me.
291 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
292 * Streamline sed()'s prototype and declare
293 * the header parse and add structures static.
295 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
296 * - Mark csp as immutable for send_crunch_response().
297 * - Fix comment spelling.
299 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
300 * Factor update_server_headers() out of sed(), ditch the
301 * first_run hack and make server_patterns_light static.
303 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
304 * Don't provide get_header() with the whole client state
305 * structure when it only needs access to csp->iob.
307 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
308 * Make the "read the rest of the headers" loop a bit more readable.
310 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
311 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
312 * - Remove useless comment.
314 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
315 * Fix comment grammar.
317 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
318 * Remove the pointless buffer in client_protocol_is_unsupported().
320 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
321 * Least-effort fix for bug #1821930 (reported by Lee):
322 * If the response doesn't look like HTTP,
323 * tell the client and log the problem.
325 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
326 * Don't pass the whole csp structure to flush_socket()
327 * when it only needs a file descriptor and a buffer.
329 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
330 * Remove kill-popups action.
332 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
333 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
335 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
336 * Remove the treat-forbidden-connects-like-blocks action. We now
337 * use the "blocked" page for forbidden CONNECT requests by default.
339 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
340 * Also use shiny new connect_port_is_forbidden() in jcc.c.
342 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
343 * Rename url_actions() to get_url_actions() and let it
344 * use the standard parameter ordering.
346 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
347 * Let forward_url() use the standard parameter ordering
348 * and mark its second parameter immutable.
350 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
351 * Remove the "Listening ... for local connections only" log message.
352 * Whether or not remote connections are able to reach Privoxy is up
353 * to the operating system.
355 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
356 * Prevent the log messages for CONNECT requests to unacceptable
357 * ports from printing the limit-connect argument as [null] if
358 * limit-connect hasn't been explicitly enabled.
360 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
361 * Make sure all console-mode apps get a usage() instance
363 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
364 * Reword NO_SERVER_DATA_RESPONSE to make it harder
365 * to misunderstand what the message is all about.
367 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
368 * Unbreak trustfile which previously didn't work without
369 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
371 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
372 * Plug memory leak. Spotted by Valgrind, triggered by
373 * Privoxy-Regression-Test feeding proxyfuzz.py.
375 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
376 * In the HTTP snipplets, refer to the client as client.
378 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
379 * Emit a log message when activating the MS IIS5 hack.
381 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
382 * Log the "weak randomization factor" warning only
383 * once for mingw32 and provide some more details.
385 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
386 * Initialize log module after initializing mutexes, future
387 * deadlocks in that code should now work cross-platform.
389 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
390 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
391 * Reported by Bert van Leeuwen in BR#1818808.
393 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
394 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
396 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
397 * When in daemon mode, close stderr after the configuration file has been
398 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
400 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
401 * - Fix a crash when parsing invalid requests whose first header
402 * is rejected by get_header(). Regression (re?)introduced
403 * in r1.143 by yours truly.
404 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
405 * client_cookie_adder() to make sure send-vanilla-wafer can be
406 * controlled through tags (and thus regression-tested).
408 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
409 * - Move get_filter_function() from jcc.c to filters.c
410 * so the filter functions can be static.
411 * - Don't bother filtering body-less responses.
413 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
414 * Execute content filters through execute_content_filter().
416 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
417 * Initialize req to NULL to make sure it's defined if the
418 * first read_socket() call fails. Reported by icmp30.
420 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
421 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
422 * extensive comments moved to user manual.
424 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
425 * Don't crash if a broken header filter wiped out the request line.
427 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
428 * Fix byte_count calculation in case of flushes
429 * and don't parse the server headers a second time.
431 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
432 * - If there's a connection problem after we already forwarded
433 * parts of the original content, just hang up. Fixes BR#1776724.
434 * - Fix warnings about unused code on mingw32.
435 * - In case of flushes, calculate the byte count
436 * less incorrectly (I think).
438 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
439 * Add some more prototypes for static functions.
441 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
442 * Comment out request_contains_null_bytes() until it's used again.
444 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
445 * #1763173 from Stefan Huehner: s@const static@static const@
446 * and declare some more functions static.
448 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
449 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
450 * and explain why they were rejected in the first place.
451 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
452 * CONNECT requests. The request line was missing.
453 * - Add two more XXX reminders as we don't have enough already.
455 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
456 * As Hal noticed, checking dispatch_cgi() as the last cruncher
457 * looks like a bug if CGI requests are blocked unintentionally,
458 * so don't do it unless the user enabled the new config option
459 * "allow-cgi-request-crunching".
461 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
462 * - Allow to rewrite the request destination behind the client's back.
463 * - Turn the weird-looking unconditional for loop that
464 * reads the client request into a conditional while loop.
465 * Move the stuff that only runs once out of the loop.
466 * - Move parts of chat(), server_content_type() and the
467 * necessary stuff to fix BR#1750917 into get_filter_function().
469 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
470 * Temporary workaround for BR#1730105.
472 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
473 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
474 * deadlocks and crashes on OpenBSD and possibly other OS with neither
475 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
476 * Thanks to Ralf Horstmann for report and solution.
478 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
479 * Add forward-override{} to change the forwarding settings through
480 * action sections. This is mainly interesting to forward different
481 * clients differently (for example based on User-Agent or request
484 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
485 * - Let usage() mention the --chroot parameter.
486 * - Use read_socket() consistently and always leave
487 * the last buffer byte alone, even in cases where
488 * null termination (currently) doesn't matter.
490 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
491 * - Fix config file loading on Unix if no config file is specified.
492 * Since r1.97 Privoxy would always interpret the last argument as
493 * config file, even if it's a valid command line option.
494 * - Abort in case of unrecognized command line options. Closes #1719696.
495 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
496 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
498 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
499 * - Don't rerun crunchers that only depend on the request URL.
500 * - Don't count redirects and CGI requests as "blocked requests".
502 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
503 * Support crunching based on tags created by server-header taggers.
505 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
506 * Make HTTP snippets static (again). Add a Content-Type for those
507 * with content so the browser doesn't guess it based on the URL.
509 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
510 * Move crunching and request line rebuilding out of chat().
512 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
513 * Introduce tags as alternative way to specify which
514 * actions apply to a request. At the moment tags can be
515 * created based on client and server headers.
517 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
518 * Don't CLF-log CONNECT requests twice.
520 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
521 * Log the source address for ACL-related connection drops.
523 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
524 * New config option: enforce-blocks.
526 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
527 * - Move null byte check into separate function.
528 * - Don't confuse the client with error pages
529 * if a CONNECT request was already confirmed.
531 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
532 * Speed up NULL byte escaping and only log the complete
533 * NULL byte requests with header debugging enabled.
535 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
536 * Answer requests that contain NULL bytes with
537 * a custom response instead of waiting for more
538 * data until the client eventually hangs up.
540 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
541 * - Move delivery and logging of crunched responses
542 * from chat() into send_crunch_response().
543 * - Display the reason for generating http_responses.
544 * - Log the content length for LOG_LEVEL_CLF correctly
545 * (still incorrect for some fixed responses).
546 * - Reword an incorrect comment about
547 * treat-forbidden-connects-like-blocks violating
549 * - Add some log messages.
551 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
552 * Move mutex initialization into separate
553 * function and exit in case of errors.
555 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
556 * - Start to reduce chat()'s line count and move
557 * parts of it into separate functions.
558 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
560 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
561 * - Add Proxy-Agent header to HTTP snippets that are
562 * supposed to reach HTTP clients only.
563 * - Made a few CONNECT log messages more descriptive.
564 * - Catch completely empty server responses (as seen
565 * with Tor's fake ".noconnect" top level domain).
566 * - Use shiny new "forwarding-failed" template for socks errors.
568 * Revision 1.118 2007/01/07 07:43:43 joergs
569 * AmigaOS4 support added.
571 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
572 * Added config option accept-intercepted-requests
573 * and disabled it by default.
575 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
576 * Reverted parts of my last commit
577 * to keep error handling working.
579 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
580 * Fixed gcc43 conversion warnings.
582 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
583 * Fix -pedantic ISO C warning about converting
584 * from function pointer to object pointer.
586 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
587 * Silence compiler warning I introduced with my last commit.
589 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
590 * Mutex protect rand() if POSIX threading
591 * is used, warn the user if that's not possible
592 * and stop using it on _WIN32 where it could
595 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
596 * Don't prevent core dumps by catching SIGABRT.
597 * It's rude and makes debugging unreasonable painful.
599 * Revision 1.110 2006/12/13 14:52:53 etresoft
600 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
602 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
603 * Privoxy is now able to run as intercepting
604 * proxy in combination with any packet filter
605 * that does the port redirection. The destination
606 * is extracted from the "Host:" header which
607 * should be available for nearly all requests.
609 * Moved HTTP snipplets into jcc.c.
610 * Added error message for gopher proxy requests.
612 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
613 * Only unlink the pidfile if it's actually used.
615 * Change order of interception checks to make
616 * it possible to block or redirect requests for
619 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
620 * Make pthread mutex locking more generic. Instead of
621 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
622 * and use mutex locking unless there is an _r function
623 * available. Better safe than sorry.
625 * Fixes "./configure --disable-pthread" and should result
626 * in less threading-related problems on pthread-using platforms,
627 * but it still doesn't fix BR#1122404.
629 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
630 * Move pthread.h inclusion from jcc.c to jcc.h.
631 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
633 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
634 * Don't exit after receiving the second SIGHUP on Solaris.
636 * Fixes BR 1052235, but the same problem may exist on other
637 * systems. Once 3.0.6 is out we should use sigset()
638 * where available and see if it breaks anything.
640 * Revision 1.104 2006/09/23 13:26:38 roro
641 * Replace TABs by spaces in source code.
643 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
644 * Fix +redirect{}. Didn't work with -fast-redirects.
646 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
647 * Respond with 400 and a short text message
648 * if the client tries to use Privoxy as FTP proxy.
650 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
651 * Make number of retries in case of forwarded-connect problems
652 * a config file option (forwarded-connect-retries) and use 0 as
655 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
656 * Set random(3) seed.
658 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
659 * Follow the OpenBSD port's lead and protect the resolve
660 * functions on OpenBSD as well.
662 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
663 * --user fix. Only use the user as group if no group is specified.
664 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
666 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
667 * Windows service (re-)integration
671 * --install[:service_name]
672 * --uninstall[:service_name]
675 * They work as follows:
676 * --install will create a service for you and then terminate.
677 * By default the service name will be "privoxy" (without the quotes).
678 * However you can run multiple services if you wish, just by adding
679 * a colon and then a name (no spaces).
681 * --uninstall follows the exact same rules a --install.
683 * --service is used when the program is executed by the service
684 * control manager, and in normal circumstances would never be
685 * used as a command line argument.
687 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
688 * Windows service integration
690 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
691 * Incorporate Fabian Keil's patch work:
692 http://www.fabiankeil.de/sourcecode/privoxy/
694 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
695 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
696 * with what was really the latest development (the v_3_0_branch branch)
698 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
699 * Thanks to Jindrich Makovicka for a race condition fix for the log
700 * file. The race condition remains for non-pthread implementations.
701 * Reference patch #1175720.
703 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
704 * Add the ability to check jpeg images for invalid
705 * lengths of comment blocks. Defensive strategy
706 * against the exploit:
707 * Microsoft Security Bulletin MS04-028
708 * Buffer Overrun in JPEG Processing (GDI+) Could
709 * Allow Code Execution (833987)
710 * Enabled with +inspect-jpegs in actions files.
712 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
713 * - Fixed usage info for non-unix platforms
714 * - Fixed small cmdline parsing bug
716 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
717 * Diagnostics: Now preserve the returncode of pthread_create
718 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
720 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
721 * No longer ignore SIGCHLD. Fixes bug #769381
723 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
724 * Close jarfile on graceful exit, remove stray line
726 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
727 * Cosmetics: Killed a warning, a typo and an allocation left at exit
729 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
730 * No longer rely on non-POSIX.1 extensions of getcwd().
733 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
734 * Replaced setenv() by posix-compliant putenv()
735 * Thanks to Neil McCalden (nmcc AT users.sf.net).
737 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
738 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
740 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
741 * Clean-up and extension of improvements for forked mode:
742 * - Child's return code now consists of flags RC_FLAG_*
743 * - Reporting toggle to parent now properly #ifdef'ed
744 * - Children now report blocking to parent. This enables
745 * statistics in forked mode
747 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
748 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
749 * Children now report having been toggled through _exit(17), parents
750 * watch for that code and toggle themselves if found.
752 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
753 * Wrapping all *_r functions (the non-_r versions of them) with
754 * mutex semaphores for OSX. Hopefully this will take care of all
755 * of those pesky crash reports.
757 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
758 * Fixed two mostly harmless mem leaks
760 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
761 * Fix: Head of global clients list now initialized to NULL
763 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
764 * Added basic support for OPTIONS and TRACE HTTP methods:
765 * - New interceptor direct_response() added in chat().
766 * - sed() moved to earlier in the process, so that the
767 * Host: header is evaluated before actions and forwarding
770 * Revision 1.92 2002/05/08 16:00:46 oes
771 * Chat's buffer handling:
772 * - Fixed bug with unchecked out-of-mem conditions
773 * while reading client request & server headers
774 * - No longer predict if the buffer limit will be exceeded
775 * in the next read -- check add_to_iob's new
776 * return code. If buffer couldn't be extended
777 * (policy or out-of-mem) while
778 * - reading from client: abort
779 * - reading server headers: send error page
780 * - buffering server body for filter: flush,
781 * and if that fails: send error page
783 * Revision 1.91 2002/04/08 20:35:58 swa
786 * Revision 1.90 2002/04/02 14:57:28 oes
787 * Made sending wafers independent of FEATURE_COOKIE_JAR
789 * Revision 1.89 2002/03/31 17:18:59 jongfoster
790 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
792 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
793 * More compiler warning message maintenance
795 * Revision 1.87 2002/03/26 22:29:54 swa
796 * we have a new homepage!
798 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
799 * Workaround for closing the jarfile before load_config() comes around again
801 * Revision 1.85 2002/03/24 15:23:33 jongfoster
804 * Revision 1.84 2002/03/24 13:25:43 swa
805 * name change related issues
807 * Revision 1.83 2002/03/16 23:54:06 jongfoster
808 * Adding graceful termination feature, to help look for memory leaks.
809 * If you enable this (which, by design, has to be done by hand
810 * editing config.h) and then go to http://i.j.b/die, then the program
811 * will exit cleanly after the *next* request. It should free all the
812 * memory that was used.
814 * Revision 1.82 2002/03/13 00:27:05 jongfoster
817 * Revision 1.81 2002/03/12 01:42:50 oes
818 * Introduced modular filters
820 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
821 * OS/2 port maintenance:
822 * - Fixed EMX build - it had decayed a little
823 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
824 * substituted a memset for now.
826 * Revision 1.79 2002/03/09 20:03:52 jongfoster
827 * - Making various functions return int rather than size_t.
828 * (Undoing a recent change). Since size_t is unsigned on
829 * Windows, functions like read_socket that return -1 on
830 * error cannot return a size_t.
832 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
833 * crashes, and also frequently caused JB to jump to 100%
834 * CPU and stay there. (Because it thought it had just
835 * read ((unsigned)-1) == 4Gb of data...)
837 * - The signature of write_socket has changed, it now simply
838 * returns success=0/failure=nonzero.
840 * - Trying to get rid of a few warnings --with-debug on
841 * Windows, I've introduced a new type "jb_socket". This is
842 * used for the socket file descriptors. On Windows, this
843 * is SOCKET (a typedef for unsigned). Everywhere else, it's
844 * an int. The error value can't be -1 any more, so it's
845 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
846 * Windows it maps to the #define INVALID_SOCKET.)
848 * - The signature of bind_port has changed.
850 * Revision 1.78 2002/03/08 21:35:04 oes
851 * Added optional group supplement to --user option. Will now use default group of user if no group given
853 * Revision 1.77 2002/03/07 03:52:06 oes
854 * - Fixed compiler warnings etc
855 * - Improved handling of failed DNS lookups
857 * Revision 1.76 2002/03/06 22:54:35 jongfoster
858 * Automated function-comment nitpicking.
860 * Revision 1.75 2002/03/06 10:02:19 oes
861 * Fixed stupid bug when --user was not given
863 * Revision 1.74 2002/03/06 00:49:31 jongfoster
864 * Fixing warning on Windows
865 * Making #ifdefs that refer to the same variable consistently
866 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
868 * Revision 1.73 2002/03/05 23:57:30 hal9
869 * Stray character 's' on line 1618 was breaking build.
871 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
872 * - Re-enable OS/2 building after new parms were added
873 * - Fix false out of memory report when resolving CGI templates when no IP
874 * address is available of failed attempt (a la no such domain)
876 * Revision 1.71 2002/03/05 18:13:56 oes
877 * Added --user option
879 * Revision 1.70 2002/03/05 04:52:42 oes
880 * Deleted non-errlog debugging code
882 * Revision 1.69 2002/03/04 23:50:00 jongfoster
883 * Splitting off bind_port() call into bind_port_helper(), with
886 * Revision 1.68 2002/03/04 20:17:32 oes
889 * Revision 1.67 2002/03/04 18:18:57 oes
890 * - Removed _DEBUG mode
891 * - Cleand up cmdline parsing
892 * - Introduced --no-daemon, --pidfile options
893 * - Cleaned up signal handling:
894 * - Terminate cleanly on INT, TERM and ABRT
895 * - Schedule logfile for re-opening on HUP
896 * - Ignore CHLD and PIPE
897 * - Leave the rest with their default handlers
898 * - Uniform handler registration
899 * - Added usage() function
900 * - Played styleguide police
902 * Revision 1.66 2002/03/03 15:06:55 oes
903 * Re-enabled automatic config reloading
905 * Revision 1.65 2002/03/03 14:49:11 oes
906 * Fixed CLF logging: Now uses client's original HTTP request
908 * Revision 1.64 2002/03/03 09:18:03 joergs
909 * Made jumbjuster work on AmigaOS again.
911 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
912 * Clean up a little CRLF unpleasantness that suddenly appeared
914 * Revision 1.62 2002/02/20 23:17:23 jongfoster
915 * Detecting some out-of memory conditions and exiting with a log message.
917 * Revision 1.61 2002/01/17 21:01:52 jongfoster
918 * Moving all our URL and URL pattern parsing code to urlmatch.c.
920 * Revision 1.60 2001/12/30 14:07:32 steudten
921 * - Add signal handling (unix)
922 * - Add SIGHUP handler (unix)
923 * - Add creation of pidfile (unix)
924 * - Add action 'top' in rc file (RH)
925 * - Add entry 'SIGNALS' to manpage
926 * - Add exit message to logfile (unix)
928 * Revision 1.59 2001/12/13 14:07:18 oes
929 * Fixed Bug: 503 error page now sent OK
931 * Revision 1.58 2001/11/30 23:37:24 jongfoster
932 * Renaming the Win32 config file to config.txt - this is almost the
933 * same as the corresponding UNIX name "config"
935 * Revision 1.57 2001/11/16 00:47:43 jongfoster
936 * Changing the tty-disconnection code to use setsid().
938 * Revision 1.56 2001/11/13 20:20:54 jongfoster
939 * Tabs->spaces, fixing a bug with missing {} around an if()
941 * Revision 1.55 2001/11/13 20:14:53 jongfoster
942 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
944 * Revision 1.54 2001/11/07 00:03:14 steudten
945 * Give reliable return value if an error
946 * occurs not just 0 with new daemon mode.
948 * Revision 1.53 2001/11/05 21:41:43 steudten
949 * Add changes to be a real daemon just for unix os.
950 * (change cwd to /, detach from controlling tty, set
951 * process group and session leader to the own process.
953 * Add some fatal-error log message for failed malloc().
954 * Add '-d' if compiled with 'configure --with-debug' to
955 * enable debug output.
957 * Revision 1.52 2001/10/26 20:11:20 jongfoster
958 * Fixing type mismatch
960 * Revision 1.51 2001/10/26 17:38:28 oes
963 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
964 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
965 * threads to call select() simultaneously. So, it's time to do a real, live,
966 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
967 * (native). Both versions will work, but using __OS2__ offers multi-threading.
969 * Revision 1.49 2001/10/23 21:41:35 jongfoster
970 * Added call to initialize the (statically-allocated of course)
971 * "out of memory" CGI response.
973 * Revision 1.48 2001/10/10 19:56:46 jongfoster
974 * Moving some code that wasn't cookie-related out of an #ifdef
977 * Revision 1.47 2001/10/10 16:44:36 oes
978 * Added CONNECT destination port limitation check
980 * Revision 1.46 2001/10/08 15:17:41 oes
981 * Re-enabled SSL forwarding
983 * Revision 1.45 2001/10/07 15:42:11 oes
984 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
986 * Moved downgrading of the HTTP version from parse_http_request to
987 * chat(), since we can't decide if it is necessary before we have
988 * determined the actions for the URL. The HTTP command is now
989 * *always* re-built so the repairs need no longer be special-cased.
991 * filter_popups now gets a csp pointer so it can raise the new
992 * CSP_FLAG_MODIFIED flag.
996 * Added configurable size limit for the IOB. If the IOB grows so
997 * large that the next read would exceed the limit, the header
998 * is generated, and the header & unfiltered buffer are flushed
999 * to the client. Chat then continues in non-buffering,
1000 * non-filtering body mode.
1002 * Revision 1.44 2001/10/02 18:13:57 oes
1005 * Revision 1.43 2001/10/02 15:32:13 oes
1006 * Moved generation of hdr
1008 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
1009 * Cleaning up 2 compiler warnings on OS/2.
1011 * Revision 1.41 2001/09/16 17:05:14 jongfoster
1012 * Removing unused #include showarg.h
1014 * Revision 1.40 2001/09/16 15:41:45 jongfoster
1015 * Fixing signed/unsigned comparison warning.
1017 * Revision 1.39 2001/09/16 13:21:27 jongfoster
1018 * Changes to use new list functions.
1020 * Revision 1.38 2001/09/16 13:01:46 jongfoster
1021 * Removing redundant function call that zeroed zalloc()'d memory.
1023 * Revision 1.37 2001/09/10 11:12:24 oes
1024 * Deleted unused variable
1026 * Revision 1.36 2001/09/10 10:56:15 oes
1027 * Silenced compiler warnings
1029 * Revision 1.35 2001/07/31 14:44:22 oes
1030 * Deleted unused size parameter from filter_popups()
1032 * Revision 1.34 2001/07/30 22:08:36 jongfoster
1033 * Tidying up #defines:
1034 * - All feature #defines are now of the form FEATURE_xxx
1035 * - Permanently turned off WIN_GUI_EDIT
1036 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
1038 * Revision 1.33 2001/07/29 19:32:00 jongfoster
1039 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
1041 * Revision 1.32 2001/07/29 18:47:05 jongfoster
1042 * Adding missing #include "loadcfg.h"
1044 * Revision 1.31 2001/07/29 12:17:48 oes
1045 * Applied pthread fix by Paul Lieverse
1047 * Revision 1.30 2001/07/25 22:57:13 jongfoster
1048 * __BEOS__ no longer overrides FEATURE_PTHREAD.
1049 * This is because FEATURE_PTHREAD will soon be widely used, so I
1050 * want to keep it simple.
1052 * Revision 1.29 2001/07/24 12:47:06 oes
1053 * Applied BeOS support update by Eugenia
1055 * Revision 1.28 2001/07/23 13:26:12 oes
1056 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
1058 * Revision 1.27 2001/07/19 19:09:47 haroon
1059 * - Added code to take care of the situation where while processing the first
1060 * server response (which includes the server header), after finding the end
1061 * of the headers we were not looking past the end of the headers for
1062 * content modification. I enabled it for filter_popups.
1063 * Someone else should look to see if other similar operations should be
1064 * done to the discarded portion of the buffer.
1066 * Note 2001/07/20: No, the other content modification mechanisms will process
1067 * the whole iob later anyway. --oes
1069 * Revision 1.26 2001/07/18 12:31:36 oes
1072 * Revision 1.25 2001/07/15 19:43:49 jongfoster
1073 * Supports POSIX threads.
1074 * Also removed some unused #includes.
1076 * Revision 1.24 2001/07/13 14:00:40 oes
1077 * - Generic content modification scheme:
1078 * Each feature has its own applicability flag that is set
1079 * from csp->action->flags.
1080 * Replaced the "filtering" int flag , by a function pointer
1081 * "content_filter" to the function that will do the content
1082 * modification. If it is != NULL, the document will be buffered
1083 * and processed through *content_filter, which must set
1084 * csp->content_length and return a modified copy of the body
1085 * or return NULL (on failiure).
1086 * - Changed csp->is_text to the more generic bitmap csp->content_type
1087 * which can currently take the valued CT_TEXT or CT_GIF
1088 * - Reformatting etc
1089 * - Removed all #ifdef PCRS
1091 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
1092 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
1093 * the `block_acl' call. This prevents a compilation error when the user
1094 * does not wish to use the "ACL" feature.
1096 * Revision 1.22 2001/06/29 21:45:41 oes
1097 * Indentation, CRLF->LF, Tab-> Space
1099 * Revision 1.21 2001/06/29 13:29:36 oes
1100 * - Cleaned up, improved comments
1101 * - Unified all possible interceptors (CGI,
1102 * block, trust, fast_redirect) in one
1103 * place, with one (CGI) answer generation
1104 * mechansim. Much clearer now.
1105 * - Removed the GIF image generation, which
1106 * is now done in filters.c:block_url()
1107 * - Made error conditions like domain lookup
1108 * failiure or (various) problems while talking
1109 * to the server use cgi.c:error_response()
1110 * instead of generating HTML/HTTP in chat() (yuck!)
1111 * - Removed logentry from cancelled commit
1113 * Revision 1.20 2001/06/09 10:55:28 jongfoster
1114 * Changing BUFSIZ ==> BUFFER_SIZE
1116 * Revision 1.19 2001/06/07 23:12:52 jongfoster
1117 * Replacing function pointer in struct gateway with a directly
1118 * called function forwarded_connect().
1119 * Replacing struct gateway with struct forward_spec
1121 * Revision 1.18 2001/06/03 19:12:16 oes
1122 * introduced new cgi handling
1124 * Revision 1.17 2001/06/01 20:07:23 jongfoster
1125 * Now uses action +image-blocker{} rather than config->tinygif
1127 * Revision 1.16 2001/06/01 18:49:17 jongfoster
1128 * Replaced "list_share" with "list" - the tiny memory gain was not
1129 * worth the extra complexity.
1131 * Revision 1.15 2001/05/31 21:24:47 jongfoster
1132 * Changed "permission" to "action" throughout.
1133 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
1134 * Moved vanilla wafer check into chat(), since we must now
1135 * decide whether or not to add it based on the URL.
1137 * Revision 1.14 2001/05/29 20:14:01 joergs
1138 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
1141 * Revision 1.13 2001/05/29 09:50:24 jongfoster
1142 * Unified blocklist/imagelist/permissionslist.
1143 * File format is still under discussion, but the internal changes
1144 * are (mostly) done.
1146 * Also modified interceptor behaviour:
1147 * - We now intercept all URLs beginning with one of the following
1148 * prefixes (and *only* these prefixes):
1150 * * http://ijbswa.sf.net/config/
1151 * * http://ijbswa.sourceforge.net/config/
1152 * - New interceptors "home page" - go to http://i.j.b/ to see it.
1153 * - Internal changes so that intercepted and fast redirect pages
1154 * are not replaced with an image.
1155 * - Interceptors now have the option to send a binary page direct
1156 * to the client. (i.e. ijb-send-banner uses this)
1157 * - Implemented show-url-info interceptor. (Which is why I needed
1158 * the above interceptors changes - a typical URL is
1159 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
1160 * The previous mechanism would not have intercepted that, and
1161 * if it had been intercepted then it then it would have replaced
1162 * it with an image.)
1164 * Revision 1.12 2001/05/27 22:17:04 oes
1166 * - re_process_buffer no longer writes the modified buffer
1167 * to the client, which was very ugly. It now returns the
1168 * buffer, which it is then written by chat.
1170 * - content_length now adjusts the Content-Length: header
1171 * for modified documents rather than crunch()ing it.
1172 * (Length info in csp->content_length, which is 0 for
1173 * unmodified documents)
1175 * - For this to work, sed() is called twice when filtering.
1177 * Revision 1.11 2001/05/26 17:27:53 jongfoster
1178 * Added support for CLF and fixed LOG_LEVEL_LOG.
1179 * Also did CRLF->LF fix of my previous patch.
1181 * Revision 1.10 2001/05/26 15:26:15 jongfoster
1182 * ACL feature now provides more security by immediately dropping
1183 * connections from untrusted hosts.
1185 * Revision 1.9 2001/05/26 00:28:36 jongfoster
1186 * Automatic reloading of config file.
1187 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
1188 * Most of the global variables have been moved to a new
1189 * struct configuration_spec, accessed through csp->config->globalname
1190 * Most of the globals remaining are used by the Win32 GUI.
1192 * Revision 1.8 2001/05/25 22:43:18 jongfoster
1193 * Fixing minor memory leak and buffer overflow.
1195 * Revision 1.7 2001/05/25 22:34:30 jongfoster
1198 * Revision 1.6 2001/05/23 00:13:58 joergs
1199 * AmigaOS support fixed.
1201 * Revision 1.5 2001/05/22 18:46:04 oes
1203 * - Enabled filtering banners by size rather than URL
1204 * by adding patterns that replace all standard banner
1205 * sizes with the "Junkbuster" gif to the re_filterfile
1207 * - Enabled filtering WebBugs by providing a pattern
1208 * which kills all 1x1 images
1210 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
1211 * which is selected by the (nonstandard and therefore
1212 * capital) letter 'U' in the option string.
1213 * It causes the quantifiers to be ungreedy by default.
1214 * Appending a ? turns back to greedy (!).
1216 * - Added a new interceptor ijb-send-banner, which
1217 * sends back the "Junkbuster" gif. Without imagelist or
1218 * MSIE detection support, or if tinygif = 1, or the
1219 * URL isn't recognized as an imageurl, a lame HTML
1220 * explanation is sent instead.
1222 * - Added new feature, which permits blocking remote
1223 * script redirects and firing back a local redirect
1225 * The feature is conditionally compiled, i.e. it
1226 * can be disabled with --disable-fast-redirects,
1227 * plus it must be activated by a "fast-redirects"
1228 * line in the config file, has its own log level
1229 * and of course wants to be displayed by show-proxy-args
1230 * Note: Boy, all the #ifdefs in 1001 locations and
1231 * all the fumbling with configure.in and acconfig.h
1232 * were *way* more work than the feature itself :-(
1234 * - Because a generic redirect template was needed for
1235 * this, tinygif = 3 now uses the same.
1237 * - Moved GIFs, and other static HTTP response templates
1240 * - Some minor fixes
1242 * - Removed some >400 CRs again (Jon, you really worked
1245 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1246 * Made failure to bind() a fatal error.
1248 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1249 * Version 2.9.4 checkin.
1250 * - Merged popupfile and cookiefile, and added control over PCRS
1251 * filtering, in new "permissionsfile".
1252 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1253 * file error you now get a message box (in the Win32 GUI) rather
1254 * than the program exiting with no explanation.
1255 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1257 * - Removed tabs from "config"
1258 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1259 * - Bumped up version number.
1261 * Revision 1.2 2001/05/17 22:34:44 oes
1262 * - Added hint on GIF char array generation to jcc.c
1263 * - Cleaned CRLF's from the sources and related files
1264 * - Repaired logging for REF and FRC
1266 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1267 * Initial import of version 2.9.3 source tree
1270 *********************************************************************/
1276 #include <sys/types.h>
1285 # ifndef FEATURE_PTHREAD
1289 # include <windows.h>
1290 # include <process.h>
1291 # endif /* ndef FEATURE_PTHREAD */
1294 # ifndef _WIN_CONSOLE
1295 # include "w32log.h"
1296 # endif /* ndef _WIN_CONSOLE */
1297 # include "w32svrapi.h"
1299 #else /* ifndef _WIN32 */
1301 # if !defined (__OS2__)
1302 # include <unistd.h>
1303 # include <sys/wait.h>
1304 # endif /* ndef __OS2__ */
1305 # include <sys/time.h>
1306 # include <sys/stat.h>
1307 # include <sys/ioctl.h>
1310 #include <sys/termios.h>
1318 # include <signal.h>
1321 # include <socket.h> /* BeOS has select() for sockets only. */
1322 # include <OS.h> /* declarations for threads and stuff. */
1325 # if defined(__EMX__) || defined(__OS2__)
1326 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1331 #define bzero(B,N) memset(B,0x00,n)
1335 # include <select.h>
1340 #include "project.h"
1343 #include "filters.h"
1344 #include "loaders.h"
1345 #include "parsers.h"
1346 #include "miscutil.h"
1348 #include "jbsockets.h"
1349 #include "gateway.h"
1350 #include "actions.h"
1352 #include "loadcfg.h"
1353 #include "urlmatch.h"
1355 const char jcc_h_rcs[] = JCC_H_VERSION;
1356 const char project_h_rcs[] = PROJECT_H_VERSION;
1359 struct client_state clients[1];
1360 struct file_list files[1];
1362 #ifdef FEATURE_STATISTICS
1363 int urls_read = 0; /* total nr of urls read inc rejected */
1364 int urls_rejected = 0; /* total nr of urls rejected */
1365 #endif /* def FEATURE_STATISTICS */
1367 #ifdef FEATURE_GRACEFUL_TERMINATION
1368 int g_terminate = 0;
1371 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1372 static void sig_handler(int the_signal);
1374 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1375 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1376 static jb_err get_server_headers(struct client_state *csp);
1377 static const char *crunch_reason(const struct http_response *rsp);
1378 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1379 static char *get_request_line(struct client_state *csp);
1380 static jb_err receive_client_request(struct client_state *csp);
1381 static jb_err parse_client_request(struct client_state *csp);
1382 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1383 static jb_err change_request_destination(struct client_state *csp);
1384 static void chat(struct client_state *csp);
1385 static void serve(struct client_state *csp);
1386 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1387 static void usage(const char *myname);
1389 static void initialize_mutexes(void);
1390 static jb_socket bind_port_helper(struct configuration_spec *config);
1391 static void listen_loop(void);
1394 void serve(struct client_state *csp);
1395 #else /* ifndef AMIGA */
1396 static void serve(struct client_state *csp);
1397 #endif /* def AMIGA */
1400 static int32 server_thread(void *data);
1401 #endif /* def __BEOS__ */
1404 #define sleep(N) Sleep(((N) * 1000))
1408 #define sleep(N) DosSleep(((N) * 100))
1411 #ifdef MUTEX_LOCKS_AVAILABLE
1413 * XXX: Does the locking stuff really belong in this file?
1415 privoxy_mutex_t log_mutex;
1416 privoxy_mutex_t log_init_mutex;
1417 privoxy_mutex_t connection_reuse_mutex;
1419 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1420 privoxy_mutex_t resolver_mutex;
1421 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1423 #ifndef HAVE_GMTIME_R
1424 privoxy_mutex_t gmtime_mutex;
1425 #endif /* ndef HAVE_GMTIME_R */
1427 #ifndef HAVE_LOCALTIME_R
1428 privoxy_mutex_t localtime_mutex;
1429 #endif /* ndef HAVE_GMTIME_R */
1432 privoxy_mutex_t rand_mutex;
1433 #endif /* ndef HAVE_RANDOM */
1435 #endif /* def MUTEX_LOCKS_AVAILABLE */
1438 const char *basedir = NULL;
1439 const char *pidfile = NULL;
1440 static int received_hup_signal = 0;
1441 #endif /* defined unix */
1443 /* HTTP snipplets. */
1444 static const char CSUCCEED[] =
1445 "HTTP/1.0 200 Connection established\r\n"
1446 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1448 static const char CHEADER[] =
1449 "HTTP/1.0 400 Invalid header received from client\r\n"
1450 "Proxy-Agent: Privoxy " VERSION "\r\n"
1451 "Content-Type: text/plain\r\n"
1452 "Connection: close\r\n\r\n"
1453 "Invalid header received from client.\r\n";
1455 static const char FTP_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 FTP.\r\n";
1461 static const char GOPHER_RESPONSE[] =
1462 "HTTP/1.0 400 Invalid request received from client\r\n"
1463 "Content-Type: text/plain\r\n"
1464 "Connection: close\r\n\r\n"
1465 "Invalid request. Privoxy doesn't support gopher.\r\n";
1467 /* XXX: should be a template */
1468 static const char MISSING_DESTINATION_RESPONSE[] =
1469 "HTTP/1.0 400 Bad request received from client\r\n"
1470 "Proxy-Agent: Privoxy " VERSION "\r\n"
1471 "Content-Type: text/plain\r\n"
1472 "Connection: close\r\n\r\n"
1473 "Bad request. Privoxy was unable to extract the destination.\r\n";
1475 /* XXX: should be a template */
1476 static const char NO_SERVER_DATA_RESPONSE[] =
1477 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1478 "Proxy-Agent: Privoxy " VERSION "\r\n"
1479 "Content-Type: text/plain\r\n"
1480 "Connection: close\r\n\r\n"
1481 "Empty server or forwarder response.\r\n"
1482 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1484 /* XXX: should be a template */
1485 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1486 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1487 "Proxy-Agent: Privoxy " VERSION "\r\n"
1488 "Content-Type: text/plain\r\n"
1489 "Connection: close\r\n\r\n"
1490 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1493 /* XXX: should be a template */
1494 static const char NULL_BYTE_RESPONSE[] =
1495 "HTTP/1.0 400 Bad request received from client\r\n"
1496 "Proxy-Agent: Privoxy " VERSION "\r\n"
1497 "Content-Type: text/plain\r\n"
1498 "Connection: close\r\n\r\n"
1499 "Bad request. Null byte(s) before end of request.\r\n";
1502 /* XXX: should be a template */
1503 static const char MESSED_UP_REQUEST_RESPONSE[] =
1504 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1505 "Proxy-Agent: Privoxy " VERSION "\r\n"
1506 "Content-Type: text/plain\r\n"
1507 "Connection: close\r\n\r\n"
1508 "Bad request. Messed up with header filters.\r\n";
1510 static const char TOO_MANY_CONNECTIONS_RESPONSE[] =
1511 "HTTP/1.0 503 Too many open connections\r\n"
1512 "Proxy-Agent: Privoxy " VERSION "\r\n"
1513 "Content-Type: text/plain\r\n"
1514 "Connection: close\r\n\r\n"
1515 "Maximum number of open connections reached.\r\n";
1517 /* XXX: should be a template */
1518 static const char CONNECTION_TIMEOUT_RESPONSE[] =
1519 "HTTP/1.0 504 Connection timeout\r\n"
1520 "Proxy-Agent: Privoxy " VERSION "\r\n"
1521 "Content-Type: text/plain\r\n"
1522 "Connection: close\r\n\r\n"
1523 "The connection timed out.\r\n";
1525 /* A function to crunch a response */
1526 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1528 /* Crunch function flags */
1529 #define CF_NO_FLAGS 0
1530 /* Cruncher applies to forced requests as well */
1531 #define CF_IGNORE_FORCE 1
1532 /* Crunched requests are counted for the block statistics */
1533 #define CF_COUNT_AS_REJECT 2
1535 /* A crunch function and its flags */
1538 const crunch_func_ptr cruncher;
1542 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1544 /* Complete list of cruncher functions */
1545 static const struct cruncher crunchers_all[] = {
1546 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1547 { block_url, CF_COUNT_AS_REJECT },
1548 #ifdef FEATURE_TRUST
1549 { trust_url, CF_COUNT_AS_REJECT },
1550 #endif /* def FEATURE_TRUST */
1551 { redirect_url, CF_NO_FLAGS },
1552 { dispatch_cgi, CF_IGNORE_FORCE},
1556 /* Light version, used after tags are applied */
1557 static const struct cruncher crunchers_light[] = {
1558 { block_url, CF_COUNT_AS_REJECT },
1559 { redirect_url, CF_NO_FLAGS },
1565 * XXX: Don't we really mean
1571 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1572 /*********************************************************************
1574 * Function : sig_handler
1576 * Description : Signal handler for different signals.
1577 * Exit gracefully on TERM and INT
1578 * or set a flag that will cause the errlog
1579 * to be reopened by the main thread on HUP.
1582 * 1 : the_signal = the signal cause this function to call
1586 *********************************************************************/
1587 static void sig_handler(int the_signal)
1593 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1605 received_hup_signal = 1;
1611 * We shouldn't be here, unless we catch signals
1612 * in main() that we can't handle here!
1614 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1622 /*********************************************************************
1624 * Function : client_protocol_is_unsupported
1626 * Description : Checks if the client used a known unsupported
1627 * protocol and deals with it by sending an error
1631 * 1 : csp = Current client state (buffers, headers, etc...)
1632 * 2 : req = the first request line send by the client
1634 * Returns : TRUE if an error response has been generated, or
1635 * FALSE if the request doesn't look invalid.
1637 *********************************************************************/
1638 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1641 * If it's a FTP or gopher request, we don't support it.
1643 * These checks are better than nothing, but they might
1644 * not work in all configurations and some clients might
1645 * have problems digesting the answer.
1647 * They should, however, never cause more problems than
1648 * Privoxy's old behaviour (returning the misleading HTML
1651 * "Could not resolve http://(ftp|gopher)://example.org").
1653 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1655 const char *response = NULL;
1656 const char *protocol = NULL;
1658 if (!strncmpic(req, "GET ftp://", 10))
1660 response = FTP_RESPONSE;
1665 response = GOPHER_RESPONSE;
1666 protocol = "GOPHER";
1668 log_error(LOG_LEVEL_ERROR,
1669 "%s tried to use Privoxy as %s proxy: %s",
1670 csp->ip_addr_str, protocol, req);
1671 log_error(LOG_LEVEL_CLF,
1672 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1674 write_socket(csp->cfd, response, strlen(response));
1683 /*********************************************************************
1685 * Function : get_request_destination_elsewhere
1687 * Description : If the client's request was redirected into
1688 * Privoxy without the client's knowledge,
1689 * the request line lacks the destination host.
1691 * This function tries to get it elsewhere,
1692 * provided accept-intercepted-requests is enabled.
1694 * "Elsewhere" currently only means "Host: header",
1695 * but in the future we may ask the redirecting
1696 * packet filter to look the destination up.
1698 * If the destination stays unknown, an error
1699 * response is send to the client and headers
1700 * are freed so that chat() can return directly.
1703 * 1 : csp = Current client state (buffers, headers, etc...)
1704 * 2 : headers = a header list
1706 * Returns : JB_ERR_OK if the destination is now known, or
1707 * JB_ERR_PARSE if it isn't.
1709 *********************************************************************/
1710 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1714 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1716 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1717 " Privoxy isn't configured to accept intercepted requests.",
1718 csp->ip_addr_str, csp->http->cmd);
1719 /* XXX: Use correct size */
1720 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1721 csp->ip_addr_str, csp->http->cmd);
1723 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1724 destroy_list(headers);
1726 return JB_ERR_PARSE;
1728 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1730 /* Split the domain we just got for pattern matching */
1731 init_domain_components(csp->http);
1737 /* We can't work without destination. Go spread the news.*/
1739 req = list_to_text(headers);
1741 /* XXX: Use correct size */
1742 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1743 csp->ip_addr_str, csp->http->cmd);
1744 log_error(LOG_LEVEL_ERROR,
1745 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1746 csp->ip_addr_str, csp->http->cmd, req);
1749 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1750 destroy_list(headers);
1752 return JB_ERR_PARSE;
1755 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1756 * to get the destination IP address, use it as host directly
1757 * or do a reverse DNS lookup first.
1762 /*********************************************************************
1764 * Function : get_server_headers
1766 * Description : Parses server headers in iob and fills them
1767 * into csp->headers so that they can later be
1771 * 1 : csp = Current client state (buffers, headers, etc...)
1773 * Returns : JB_ERR_OK if everything went fine, or
1774 * JB_ERR_PARSE if the headers were incomplete.
1776 *********************************************************************/
1777 static jb_err get_server_headers(struct client_state *csp)
1779 int continue_hack_in_da_house = 0;
1782 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1787 * continue hack in da house. Ignore the ending of
1788 * this head and continue enlisting header lines.
1789 * The reason is described below.
1791 enlist(csp->headers, "");
1792 continue_hack_in_da_house = 0;
1795 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1798 * It's a bodyless continue response, don't
1799 * stop header parsing after reaching its end.
1801 * As a result Privoxy will concatenate the
1802 * next response's head and parse and deliver
1803 * the headers as if they belonged to one request.
1805 * The client will separate them because of the
1806 * empty line between them.
1808 * XXX: What we're doing here is clearly against
1809 * the intended purpose of the continue header,
1810 * and under some conditions (HTTP/1.0 client request)
1811 * it's a standard violation.
1813 * Anyway, "sort of against the spec" is preferable
1814 * to "always getting confused by Continue responses"
1815 * (Privoxy's behaviour before this hack was added)
1817 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1818 continue_hack_in_da_house = 1;
1820 else if (*header == '\0')
1823 * If the header is empty, but the Continue hack
1824 * isn't active, we can assume that we reached the
1825 * end of the buffer before we hit the end of the
1828 * Inform the caller an let it decide how to handle it.
1830 return JB_ERR_PARSE;
1833 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1836 * XXX: Should we quit the request and return a
1837 * out of memory error page instead?
1839 log_error(LOG_LEVEL_ERROR,
1840 "Out of memory while enlisting server headers. %s lost.",
1850 /*********************************************************************
1852 * Function : crunch_reason
1854 * Description : Translates the crunch reason code into a string.
1857 * 1 : rsp = a http_response
1859 * Returns : A string with the crunch reason or an error description.
1861 *********************************************************************/
1862 static const char *crunch_reason(const struct http_response *rsp)
1864 char * reason = NULL;
1866 assert(rsp != NULL);
1869 return "Internal error while searching for crunch reason";
1872 switch (rsp->reason)
1874 case RSP_REASON_UNSUPPORTED:
1875 reason = "Unsupported HTTP feature";
1877 case RSP_REASON_BLOCKED:
1880 case RSP_REASON_UNTRUSTED:
1881 reason = "Untrusted";
1883 case RSP_REASON_REDIRECTED:
1884 reason = "Redirected";
1886 case RSP_REASON_CGI_CALL:
1887 reason = "CGI Call";
1889 case RSP_REASON_NO_SUCH_DOMAIN:
1890 reason = "DNS failure";
1892 case RSP_REASON_FORWARDING_FAILED:
1893 reason = "Forwarding failed";
1895 case RSP_REASON_CONNECT_FAILED:
1896 reason = "Connection failure";
1898 case RSP_REASON_OUT_OF_MEMORY:
1899 reason = "Out of memory (may mask other reasons)";
1902 reason = "No reason recorded";
1910 /*********************************************************************
1912 * Function : send_crunch_response
1914 * Description : Delivers already prepared response for
1915 * intercepted requests, logs the interception
1916 * and frees the response.
1919 * 1 : csp = Current client state (buffers, headers, etc...)
1920 * 1 : rsp = Fully prepared response. Will be freed on exit.
1922 * Returns : Nothing.
1924 *********************************************************************/
1925 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1927 const struct http_request *http = csp->http;
1928 char status_code[4];
1930 assert(rsp != NULL);
1931 assert(rsp->head != NULL);
1936 * Not supposed to happen. If it does
1937 * anyway, treat it as an unknown error.
1939 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1940 /* return code doesn't matter */
1945 /* If rsp is still NULL, we have serious internal problems. */
1946 log_error(LOG_LEVEL_FATAL,
1947 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1951 * Extract the status code from the actual head
1952 * that was send to the client. It is the only
1953 * way to get it right for all requests, including
1954 * the fixed ones for out-of-memory problems.
1956 * A head starts like this: 'HTTP/1.1 200...'
1960 status_code[0] = rsp->head[9];
1961 status_code[1] = rsp->head[10];
1962 status_code[2] = rsp->head[11];
1963 status_code[3] = '\0';
1965 /* Write the answer to the client */
1966 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1967 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1969 /* There is nothing we can do about it. */
1970 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1973 /* Log that the request was crunched and why. */
1974 log_error(LOG_LEVEL_CRUNCH, "%s: %s", crunch_reason(rsp), http->url);
1975 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %u",
1976 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1978 /* Clean up and return */
1979 if (cgi_error_memory() != rsp)
1981 free_http_response(rsp);
1988 /*********************************************************************
1990 * Function : request_contains_null_bytes
1992 * Description : Checks for NULL bytes in the request and sends
1993 * an error message to the client if any were found.
1995 * XXX: currently not used, see comment in chat().
1998 * 1 : csp = Current client state (buffers, headers, etc...)
1999 * 2 : buf = Data from the client's request to check.
2000 * 3 : len = The data length.
2002 * Returns : TRUE if the request contained one or more NULL bytes, or
2005 *********************************************************************/
2006 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
2008 size_t c_len; /* Request lenght when treated as C string */
2010 c_len = strlen(buf);
2015 * Null byte(s) found. Log the request,
2016 * return an error response and hang up.
2018 size_t tmp_len = c_len;
2023 * Replace NULL byte(s) with '°' characters
2024 * so the request can be logged as string.
2025 * XXX: Is there a better replacement character?
2028 tmp_len += strlen(buf+tmp_len);
2029 } while (tmp_len < len);
2031 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
2032 "(length=%d, strlen=%u).", csp->ip_addr_str, len, c_len);
2033 log_error(LOG_LEVEL_HEADER,
2034 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
2036 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
2038 /* XXX: Log correct size */
2039 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2049 /*********************************************************************
2051 * Function : crunch_response_triggered
2053 * Description : Checks if the request has to be crunched,
2054 * and delivers the crunch response if necessary.
2057 * 1 : csp = Current client state (buffers, headers, etc...)
2058 * 2 : crunchers = list of cruncher functions to run
2060 * Returns : TRUE if the request was answered with a crunch response
2063 *********************************************************************/
2064 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
2066 struct http_response *rsp = NULL;
2067 const struct cruncher *c;
2070 * If CGI request crunching is disabled,
2071 * check the CGI dispatcher out of order to
2072 * prevent unintentional blocks or redirects.
2074 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
2075 && (NULL != (rsp = dispatch_cgi(csp))))
2077 /* Deliver, log and free the interception response. */
2078 send_crunch_response(csp, rsp);
2082 for (c = crunchers; c->cruncher != NULL; c++)
2085 * Check the cruncher if either Privoxy is toggled
2086 * on and the request isn't forced, or if the cruncher
2087 * applies to forced requests as well.
2089 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
2090 !(csp->flags & CSP_FLAG_FORCED)) ||
2091 (c->flags & CF_IGNORE_FORCE))
2093 rsp = c->cruncher(csp);
2096 /* Deliver, log and free the interception response. */
2097 send_crunch_response(csp, rsp);
2098 #ifdef FEATURE_STATISTICS
2099 if (c->flags & CF_COUNT_AS_REJECT)
2101 csp->flags |= CSP_FLAG_REJECTED;
2103 #endif /* def FEATURE_STATISTICS */
2114 /*********************************************************************
2116 * Function : build_request_line
2118 * Description : Builds the HTTP request line.
2120 * If a HTTP forwarder is used it expects the whole URL,
2121 * web servers only get the path.
2124 * 1 : csp = Current client state (buffers, headers, etc...)
2125 * 2 : fwd = The forwarding spec used for the request
2126 * XXX: Should use http->fwd instead.
2127 * 3 : request_line = The old request line which will be replaced.
2129 * Returns : Nothing. Terminates in case of memory problems.
2131 *********************************************************************/
2132 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
2134 struct http_request *http = csp->http;
2136 assert(http->ssl == 0);
2139 * Downgrade http version from 1.1 to 1.0
2140 * if +downgrade action applies.
2142 if ( (csp->action->flags & ACTION_DOWNGRADE)
2143 && (!strcmpic(http->ver, "HTTP/1.1")))
2146 http->ver = strdup("HTTP/1.0");
2148 if (http->ver == NULL)
2150 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
2155 * Rebuild the request line.
2157 freez(*request_line);
2158 *request_line = strdup(http->gpc);
2159 string_append(request_line, " ");
2161 if (fwd->forward_host)
2163 string_append(request_line, http->url);
2167 string_append(request_line, http->path);
2169 string_append(request_line, " ");
2170 string_append(request_line, http->ver);
2172 if (*request_line == NULL)
2174 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
2176 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
2180 /*********************************************************************
2182 * Function : change_request_destination
2184 * Description : Parse a (rewritten) request line and regenerate
2185 * the http request data.
2188 * 1 : csp = Current client state (buffers, headers, etc...)
2190 * Returns : Forwards the parse_http_request() return code.
2191 * Terminates in case of memory problems.
2193 *********************************************************************/
2194 static jb_err change_request_destination(struct client_state *csp)
2196 struct http_request *http = csp->http;
2199 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
2200 free_http_request(http);
2201 err = parse_http_request(csp->headers->first->str, http);
2202 if (JB_ERR_OK != err)
2204 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
2205 jb_err_to_string(err));
2209 /* XXX: ocmd is a misleading name */
2210 http->ocmd = strdup(http->cmd);
2211 if (http->ocmd == NULL)
2213 log_error(LOG_LEVEL_FATAL,
2214 "Out of memory copying rewritten HTTP request line");
2222 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2223 /*********************************************************************
2225 * Function : server_response_is_complete
2227 * Description : Determines whether we should stop reading
2228 * from the server socket.
2231 * 1 : csp = Current client state (buffers, headers, etc...)
2232 * 2 : content_length = Length of content received so far.
2234 * Returns : TRUE if the response is complete,
2237 *********************************************************************/
2238 static int server_response_is_complete(struct client_state *csp,
2239 unsigned long long content_length)
2241 int content_length_known = !!(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2243 if (!strcmpic(csp->http->gpc, "HEAD"))
2246 * "HEAD" implies no body, we are thus expecting
2247 * no content. XXX: incomplete "list" of methods?
2249 csp->expected_content_length = 0;
2250 content_length_known = TRUE;
2253 if (csp->http->status == 304)
2256 * Expect no body. XXX: incomplete "list" of status codes?
2258 csp->expected_content_length = 0;
2259 content_length_known = TRUE;
2262 return (content_length_known && ((0 == csp->expected_content_length)
2263 || (csp->expected_content_length <= content_length)));
2267 /*********************************************************************
2269 * Function : wait_for_alive_connections
2271 * Description : Waits for alive connections to timeout.
2277 *********************************************************************/
2278 static void wait_for_alive_connections()
2280 int connections_alive = close_unusable_connections();
2282 while (0 < connections_alive)
2284 log_error(LOG_LEVEL_CONNECT,
2285 "Waiting for %d connections to timeout.",
2288 connections_alive = close_unusable_connections();
2291 log_error(LOG_LEVEL_CONNECT, "No connections to wait for left.");
2296 /*********************************************************************
2298 * Function : save_connection_destination
2300 * Description : Remembers a connection for reuse later on.
2303 * 1 : sfd = Open socket to remember.
2304 * 2 : http = The destination for the connection.
2305 * 3 : fwd = The forwarder settings used.
2306 * 3 : server_connection = storage.
2310 *********************************************************************/
2311 void save_connection_destination(jb_socket sfd,
2312 const struct http_request *http,
2313 const struct forward_spec *fwd,
2314 struct reusable_connection *server_connection)
2316 assert(sfd != JB_INVALID_SOCKET);
2317 assert(NULL != http->host);
2318 server_connection->host = strdup(http->host);
2319 if (NULL == server_connection->host)
2321 log_error(LOG_LEVEL_FATAL, "Out of memory saving socket.");
2323 server_connection->port = http->port;
2325 assert(NULL != fwd);
2326 assert(server_connection->gateway_host == NULL);
2327 assert(server_connection->gateway_port == 0);
2328 assert(server_connection->forwarder_type == 0);
2329 assert(server_connection->forward_host == NULL);
2330 assert(server_connection->forward_port == 0);
2332 server_connection->forwarder_type = fwd->type;
2333 if (NULL != fwd->gateway_host)
2335 server_connection->gateway_host = strdup(fwd->gateway_host);
2336 if (NULL == server_connection->gateway_host)
2338 log_error(LOG_LEVEL_FATAL, "Out of memory saving gateway_host.");
2343 server_connection->gateway_host = NULL;
2345 server_connection->gateway_port = fwd->gateway_port;
2347 if (NULL != fwd->forward_host)
2349 server_connection->forward_host = strdup(fwd->forward_host);
2350 if (NULL == server_connection->forward_host)
2352 log_error(LOG_LEVEL_FATAL, "Out of memory saving forward_host.");
2357 server_connection->forward_host = NULL;
2359 server_connection->forward_port = fwd->forward_port;
2361 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2364 /*********************************************************************
2366 * Function : mark_server_socket_tainted
2368 * Description : Makes sure we don't reuse a server socket
2369 * (if we didn't read everything the server sent
2370 * us reusing the socket would lead to garbage).
2373 * 1 : csp = Current client state (buffers, headers, etc...)
2377 *********************************************************************/
2378 static void mark_server_socket_tainted(struct client_state *csp)
2380 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2382 log_error(LOG_LEVEL_CONNECT, "Unsetting keep-alive flag.");
2383 csp->flags &= ~CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE;
2387 /*********************************************************************
2389 * Function : get_request_line
2391 * Description : Read the client request line.
2394 * 1 : csp = Current client state (buffers, headers, etc...)
2396 * Returns : Pointer to request line or NULL in case of errors.
2398 *********************************************************************/
2399 static char *get_request_line(struct client_state *csp)
2401 char buf[BUFFER_SIZE];
2402 char *request_line = NULL;
2405 memset(buf, 0, sizeof(buf));
2409 if (!data_is_available(csp->cfd, csp->config->socket_timeout))
2411 log_error(LOG_LEVEL_ERROR,
2412 "Stopped waiting for the request line.");
2413 write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE,
2414 strlen(CONNECTION_TIMEOUT_RESPONSE));
2418 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2420 if (len <= 0) return NULL;
2423 * If there is no memory left for buffering the
2424 * request, there is nothing we can do but hang up
2426 if (add_to_iob(csp, buf, len))
2431 request_line = get_header(csp->iob);
2433 } while ((NULL != request_line) && ('\0' == *request_line));
2435 return request_line;
2440 /*********************************************************************
2442 * Function : receive_client_request
2444 * Description : Read the client's request (more precisely the
2445 * client headers) and answer it if necessary.
2447 * Note that since we're not using select() we could get
2448 * blocked here if a client connected, then didn't say
2452 * 1 : csp = Current client state (buffers, headers, etc...)
2454 * Returns : JB_ERR_OK, JB_ERR_PARSE or JB_ERR_MEMORY
2456 *********************************************************************/
2457 static jb_err receive_client_request(struct client_state *csp)
2459 char buf[BUFFER_SIZE];
2462 struct http_request *http;
2466 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2467 struct list header_list;
2468 struct list *headers = &header_list;
2472 memset(buf, 0, sizeof(buf));
2474 req = get_request_line(csp);
2477 return JB_ERR_PARSE;
2479 assert(*req != '\0');
2481 if (client_protocol_is_unsupported(csp, req))
2483 return JB_ERR_PARSE;
2486 #ifdef FEATURE_FORCE_LOAD
2488 * If this request contains the FORCE_PREFIX and blocks
2489 * aren't enforced, get rid of it and set the force flag.
2491 if (strstr(req, FORCE_PREFIX))
2493 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2495 log_error(LOG_LEVEL_FORCE,
2496 "Ignored force prefix in request: \"%s\".", req);
2500 strclean(req, FORCE_PREFIX);
2501 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2502 csp->flags |= CSP_FLAG_FORCED;
2505 #endif /* def FEATURE_FORCE_LOAD */
2507 err = parse_http_request(req, http);
2509 if (JB_ERR_OK != err)
2511 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2512 /* XXX: Use correct size */
2513 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2514 log_error(LOG_LEVEL_ERROR,
2515 "Couldn't parse request line received from %s: %s",
2516 csp->ip_addr_str, jb_err_to_string(err));
2518 free_http_request(http);
2519 return JB_ERR_PARSE;
2522 /* grab the rest of the client's headers */
2526 p = get_header(csp->iob);
2530 /* There are no additional headers to read. */
2537 * We didn't receive a complete header
2538 * line yet, get the rest of it.
2540 if (!data_is_available(csp->cfd, csp->config->socket_timeout))
2542 log_error(LOG_LEVEL_ERROR,
2543 "Stopped grabbing the client headers.");
2544 return JB_ERR_PARSE;
2547 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2550 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2551 destroy_list(headers);
2552 return JB_ERR_PARSE;
2555 if (add_to_iob(csp, buf, len))
2558 * If there is no memory left for buffering the
2559 * request, there is nothing we can do but hang up
2561 destroy_list(headers);
2562 return JB_ERR_MEMORY;
2568 * We were able to read a complete
2569 * header and can finaly enlist it.
2576 if (http->host == NULL)
2579 * If we still don't know the request destination,
2580 * the request is invalid or the client uses
2581 * Privoxy without its knowledge.
2583 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2586 * Our attempts to get the request destination
2587 * elsewhere failed or Privoxy is configured
2588 * to only accept proxy requests.
2590 * An error response has already been send
2591 * and we're done here.
2593 return JB_ERR_PARSE;
2598 * Determine the actions for this URL
2600 #ifdef FEATURE_TOGGLE
2601 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2603 /* Most compatible set of actions (i.e. none) */
2604 init_current_action(csp->action);
2607 #endif /* ndef FEATURE_TOGGLE */
2609 get_url_actions(csp, http);
2613 * Save a copy of the original request for logging
2615 http->ocmd = strdup(http->cmd);
2616 if (http->ocmd == NULL)
2618 log_error(LOG_LEVEL_FATAL,
2619 "Out of memory copying HTTP request line");
2621 enlist(csp->headers, http->cmd);
2623 /* Append the previously read headers */
2624 list_append_list_unique(csp->headers, headers);
2625 destroy_list(headers);
2632 /*********************************************************************
2634 * Function : parse_client_request
2636 * Description : Parses the client's request and decides what to do
2639 * Note that since we're not using select() we could get
2640 * blocked here if a client connected, then didn't say
2644 * 1 : csp = Current client state (buffers, headers, etc...)
2646 * Returns : JB_ERR_OK or JB_ERR_PARSE
2648 *********************************************************************/
2649 static jb_err parse_client_request(struct client_state *csp)
2651 struct http_request *http = csp->http;
2654 err = sed(csp, FILTER_CLIENT_HEADERS);
2655 if (JB_ERR_OK != err)
2657 /* XXX: Should be handled in sed(). */
2658 assert(err == JB_ERR_PARSE);
2659 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers.");
2661 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2663 /* Check request line for rewrites. */
2664 if ((NULL == csp->headers->first->str)
2665 || (strcmp(http->cmd, csp->headers->first->str) &&
2666 (JB_ERR_OK != change_request_destination(csp))))
2669 * A header filter broke the request line - bail out.
2671 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2672 /* XXX: Use correct size */
2673 log_error(LOG_LEVEL_CLF,
2674 "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2675 log_error(LOG_LEVEL_ERROR,
2676 "Invalid request line after applying header filters.");
2677 free_http_request(http);
2679 return JB_ERR_PARSE;
2687 /*********************************************************************
2691 * Description : Once a connection to the client has been accepted,
2692 * this function is called (via serve()) to handle the
2693 * main business of the communication. When this
2694 * function returns, the caller must close the client
2697 * FIXME: chat is nearly thousand lines long.
2701 * 1 : csp = Current client state (buffers, headers, etc...)
2703 * Returns : Nothing.
2705 *********************************************************************/
2706 static void chat(struct client_state *csp)
2708 char buf[BUFFER_SIZE];
2715 int ms_iis5_hack = 0;
2716 unsigned long long byte_count = 0;
2717 int forwarded_connect_retries = 0;
2718 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2719 const struct forward_spec *fwd;
2720 struct http_request *http;
2721 long len = 0; /* for buffer sizes (and negative error codes) */
2723 /* Function that does the content filtering for the current request */
2724 filter_function_ptr content_filter = NULL;
2726 /* Skeleton for HTTP response, if we should intercept the request */
2727 struct http_response *rsp;
2728 struct timeval timeout;
2730 memset(buf, 0, sizeof(buf));
2734 if (receive_client_request(csp) != JB_ERR_OK)
2738 if (parse_client_request(csp) != JB_ERR_OK)
2743 /* decide how to route the HTTP request */
2744 fwd = forward_url(csp, http);
2747 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2748 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2753 * build the http request to send to the server
2754 * we have to do one of the following:
2756 * create = use the original HTTP request to create a new
2757 * HTTP request that has either the path component
2758 * without the http://domainspec (w/path) or the
2759 * full orininal URL (w/url)
2760 * Note that the path and/or the HTTP version may
2761 * have been altered by now.
2763 * connect = Open a socket to the host:port of the server
2764 * and short-circuit server and client socket.
2766 * pass = Pass the request unchanged if forwarding a CONNECT
2767 * request to a parent proxy. Note that we'll be sending
2768 * the CFAIL message ourselves if connecting to the parent
2769 * fails, but we won't send a CSUCCEED message if it works,
2770 * since that would result in a double message (ours and the
2771 * parent's). After sending the request to the parent, we simply
2774 * here's the matrix:
2777 * +--------+--------+
2779 * 0 | create | connect|
2781 * Forwarding +--------+--------+
2783 * 1 | create | pass |
2785 * +--------+--------+
2789 if (http->ssl && connect_port_is_forbidden(csp))
2791 const char *acceptable_connect_ports =
2792 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2793 assert(NULL != acceptable_connect_ports);
2794 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2795 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2796 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2797 csp->action->flags |= ACTION_BLOCK;
2803 freez(csp->headers->first->str);
2804 build_request_line(csp, fwd, &csp->headers->first->str);
2808 * We have a request. Check if one of the crunchers wants it.
2810 if (crunch_response_triggered(csp, crunchers_all))
2813 * Yes. The client got the crunch response
2814 * and we are done here after cleaning up.
2816 /* XXX: why list_remove_all()? */
2817 list_remove_all(csp->headers);
2822 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2824 if (fwd->forward_host)
2826 log_error(LOG_LEVEL_CONNECT, "via [%s]:%d to: %s",
2827 fwd->forward_host, fwd->forward_port, http->hostport);
2831 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2834 /* here we connect to the server, gateway, or the forwarder */
2836 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2837 if ((csp->sfd != JB_INVALID_SOCKET)
2838 && socket_is_still_usable(csp->sfd)
2839 && connection_destination_matches(&csp->server_connection, http, fwd))
2841 log_error(LOG_LEVEL_CONNECT,
2842 "Reusing server socket %u. Opened for %s.",
2843 csp->sfd, csp->server_connection.host);
2847 if (csp->sfd != JB_INVALID_SOCKET)
2849 log_error(LOG_LEVEL_CONNECT,
2850 "Closing server socket %u. Opened for %s.",
2851 csp->sfd, csp->server_connection.host);
2852 close_socket(csp->sfd);
2853 mark_connection_closed(&csp->server_connection);
2855 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
2857 while ((csp->sfd = forwarded_connect(fwd, http, csp))
2858 && (errno == EINVAL)
2859 && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2861 log_error(LOG_LEVEL_ERROR,
2862 "failed request #%u to connect to %s. Trying again.",
2863 forwarded_connect_retries, http->hostport);
2866 if (csp->sfd == JB_INVALID_SOCKET)
2868 if (fwd->type != SOCKS_NONE)
2871 rsp = error_response(csp, "forwarding-failed", errno);
2873 else if (errno == EINVAL)
2875 rsp = error_response(csp, "no-such-domain", errno);
2879 rsp = error_response(csp, "connect-failed", errno);
2880 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2884 /* Write the answer to the client */
2887 send_crunch_response(csp, rsp);
2892 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2893 save_connection_destination(csp->sfd, http, fwd, &csp->server_connection);
2895 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
2897 hdr = list_to_text(csp->headers);
2900 /* FIXME Should handle error properly */
2901 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2903 list_remove_all(csp->headers);
2905 if (fwd->forward_host || (http->ssl == 0))
2908 * Write the client's (modified) header to the server
2909 * (along with anything else that may be in the buffer)
2911 if (write_socket(csp->sfd, hdr, strlen(hdr))
2912 || (flush_socket(csp->sfd, csp->iob) < 0))
2914 log_error(LOG_LEVEL_CONNECT,
2915 "write header to: %s failed: %E", http->hostport);
2917 rsp = error_response(csp, "connect-failed", errno);
2920 send_crunch_response(csp, rsp);
2930 * We're running an SSL tunnel and we're not forwarding,
2931 * so just send the "connect succeeded" message to the
2932 * client, flush the rest, and get out of the way.
2934 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2942 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2944 /* we're finished with the client's header */
2947 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2949 /* pass data between the client and server
2950 * until one or the other shuts down the connection.
2959 * FD_ZERO here seems to point to an errant macro which crashes.
2960 * So do this by hand for now...
2962 memset(&rfds,0x00,sizeof(fd_set));
2966 FD_SET(csp->cfd, &rfds);
2967 FD_SET(csp->sfd, &rfds);
2969 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2970 if ((csp->flags & CSP_FLAG_CHUNKED)
2971 && !(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
2972 && ((csp->iob->eod - csp->iob->cur) >= 5)
2973 && !memcmp(csp->iob->eod-5, "0\r\n\r\n", 5))
2975 log_error(LOG_LEVEL_CONNECT,
2976 "Looks like we read the last chunk together with "
2977 "the server headers. We better stop reading.");
2978 byte_count = (unsigned long long)(csp->iob->eod - csp->iob->cur);
2979 csp->expected_content_length = byte_count;
2980 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2982 if (server_body && server_response_is_complete(csp, byte_count))
2984 log_error(LOG_LEVEL_CONNECT,
2985 "Done reading from server. Expected content length: %llu. "
2986 "Actual content length: %llu. Most recently received: %d.",
2987 csp->expected_content_length, byte_count, len);
2990 * XXX: should not jump around,
2991 * chat() is complicated enough already.
2995 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2997 timeout.tv_sec = csp->config->socket_timeout;
2998 timeout.tv_usec = 0;
2999 n = select((int)maxfd+1, &rfds, NULL, NULL, &timeout);
3003 log_error(LOG_LEVEL_ERROR,
3004 "Didn't receive data in time: %s", http->url);
3005 if ((byte_count == 0) && (http->ssl == 0))
3007 write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE,
3008 strlen(CONNECTION_TIMEOUT_RESPONSE));
3010 mark_server_socket_tainted(csp);
3015 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
3016 mark_server_socket_tainted(csp);
3021 * This is the body of the browser's request,
3022 * just read and write it.
3024 * XXX: Make sure the client doesn't use pipelining
3025 * behind Privoxy's back.
3027 if (FD_ISSET(csp->cfd, &rfds))
3029 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
3033 /* XXX: not sure if this is necessary. */
3034 mark_server_socket_tainted(csp);
3035 break; /* "game over, man" */
3038 if (write_socket(csp->sfd, buf, (size_t)len))
3040 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
3041 mark_server_socket_tainted(csp);
3048 * The server wants to talk. It could be the header or the body.
3049 * If `hdr' is null, then it's the header otherwise it's the body.
3050 * FIXME: Does `hdr' really mean `host'? No.
3052 if (FD_ISSET(csp->sfd, &rfds))
3055 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
3059 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
3061 if (http->ssl && (fwd->forward_host == NULL))
3064 * Just hang up. We already confirmed the client's CONNECT
3065 * request with status code 200 and unencrypted content is
3066 * no longer welcome.
3068 log_error(LOG_LEVEL_ERROR,
3069 "CONNECT already confirmed. Unable to tell the client about the problem.");
3072 else if (byte_count)
3075 * Just hang up. We already transmitted the original headers
3076 * and parts of the original content and therefore missed the
3077 * chance to send an error message (without risking data corruption).
3079 * XXX: we could retry with a fancy range request here.
3081 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
3082 "Unable to tell the client about the problem.");
3083 mark_server_socket_tainted(csp);
3087 * XXX: Consider handling the cases above the same.
3089 mark_server_socket_tainted(csp);
3093 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3094 if (csp->flags & CSP_FLAG_CHUNKED)
3096 if ((len >= 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
3098 /* XXX: this is a temporary hack */
3099 log_error(LOG_LEVEL_CONNECT,
3100 "Looks like we reached the end of the last chunk. "
3101 "We better stop reading.");
3102 csp->expected_content_length = byte_count + (unsigned long long)len;
3103 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
3107 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
3110 * Add a trailing zero to let be able to use string operations.
3111 * XXX: do we still need this with filter_popups gone?
3116 * Normally, this would indicate that we've read
3117 * as much as the server has sent us and we can
3118 * close the client connection. However, Microsoft
3119 * in its wisdom has released IIS/5 with a bug that
3120 * prevents it from sending the trailing \r\n in
3121 * a 302 redirect header (and possibly other headers).
3122 * To work around this if we've haven't parsed
3123 * a full header we'll append a trailing \r\n
3124 * and see if this now generates a valid one.
3126 * This hack shouldn't have any impacts. If we've
3127 * already transmitted the header or if this is a
3128 * SSL connection, then we won't bother with this
3129 * hack. So we only work on partially received
3130 * headers. If we append a \r\n and this still
3131 * doesn't generate a valid header, then we won't
3132 * transmit anything to the client.
3137 if (server_body || http->ssl)
3140 * If we have been buffering up the document,
3141 * now is the time to apply content modification
3142 * and send the result to the client.
3146 p = execute_content_filter(csp, content_filter);
3148 * If the content filter fails, use the original
3149 * buffer and length.
3150 * (see p != NULL ? p : csp->iob->cur below)
3154 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
3157 if (JB_ERR_OK != update_server_headers(csp))
3159 log_error(LOG_LEVEL_FATAL,
3160 "Failed to update server headers. after filtering.");
3163 hdr = list_to_text(csp->headers);
3166 /* FIXME Should handle error properly */
3167 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3170 if (write_socket(csp->cfd, hdr, strlen(hdr))
3171 || write_socket(csp->cfd,
3172 ((p != NULL) ? p : csp->iob->cur), (size_t)csp->content_length))
3174 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
3177 mark_server_socket_tainted(csp);
3185 break; /* "game over, man" */
3189 * This is NOT the body, so
3190 * Let's pretend the server just sent us a blank line.
3192 snprintf(buf, sizeof(buf), "\r\n");
3193 len = (int)strlen(buf);
3196 * Now, let the normal header parsing algorithm below do its
3197 * job. If it fails, we'll exit instead of continuing.
3204 * If this is an SSL connection or we're in the body
3205 * of the server document, just write it to the client,
3206 * unless we need to buffer the body for later content-filtering
3208 if (server_body || http->ssl)
3213 * If there is no memory left for buffering the content, or the buffer limit
3214 * has been reached, switch to non-filtering mode, i.e. make & write the
3215 * header, flush the iob and buf, and get out of the way.
3217 if (add_to_iob(csp, buf, len))
3222 log_error(LOG_LEVEL_INFO,
3223 "Flushing header and buffers. Stepping back from filtering.");
3225 hdr = list_to_text(csp->headers);
3229 * Memory is too tight to even generate the header.
3230 * Send our static "Out-of-memory" page.
3232 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
3233 rsp = cgi_error_memory();
3234 send_crunch_response(csp, rsp);
3235 mark_server_socket_tainted(csp);
3238 hdrlen = strlen(hdr);
3240 if (write_socket(csp->cfd, hdr, hdrlen)
3241 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
3242 || (write_socket(csp->cfd, buf, (size_t)len)))
3244 log_error(LOG_LEVEL_CONNECT,
3245 "Flush header and buffers to client failed: %E");
3247 mark_server_socket_tainted(csp);
3252 * Reset the byte_count to the amount of bytes
3253 * we just flushed. len will be added a few lines below,
3254 * hdrlen doesn't matter for LOG_LEVEL_CLF.
3256 byte_count = (unsigned long long)flushed;
3258 content_filter = NULL;
3264 if (write_socket(csp->cfd, buf, (size_t)len))
3266 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
3267 mark_server_socket_tainted(csp);
3271 byte_count += (unsigned long long)len;
3276 const char *header_start;
3278 * We're still looking for the end of the server's header.
3279 * Buffer up the data we just read. If that fails, there's
3280 * little we can do but send our static out-of-memory page.
3282 if (add_to_iob(csp, buf, len))
3284 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
3285 rsp = cgi_error_memory();
3286 send_crunch_response(csp, rsp);
3287 mark_server_socket_tainted(csp);
3291 header_start = csp->iob->cur;
3293 /* Convert iob into something sed() can digest */
3294 if (JB_ERR_PARSE == get_server_headers(csp))
3299 * Well, we tried our MS IIS/5 hack and it didn't work.
3300 * The header is incomplete and there isn't anything
3301 * we can do about it.
3303 log_error(LOG_LEVEL_ERROR, "Invalid server headers. "
3304 "Applying the MS IIS5 hack didn't help.");
3305 log_error(LOG_LEVEL_CLF,
3306 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3307 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3308 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3309 mark_server_socket_tainted(csp);
3315 * Since we have to wait for more from the server before
3316 * we can parse the headers we just continue here.
3318 long header_offset = csp->iob->cur - header_start;
3319 assert(csp->iob->cur >= header_start);
3320 byte_count += (unsigned long long)(len - header_offset);
3321 log_error(LOG_LEVEL_CONNECT, "Continuing buffering headers. "
3322 "byte_count: %llu. header_offset: %d. len: %d.",
3323 byte_count, header_offset, len);
3328 /* Did we actually get anything? */
3329 if (NULL == csp->headers->first)
3331 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
3332 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3333 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
3334 free_http_request(http);
3335 mark_server_socket_tainted(csp);
3339 assert(csp->headers->first->str);
3341 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
3342 strncmpic(csp->headers->first->str, "ICY", 3))
3345 * It doesn't look like a HTTP (or Shoutcast) response:
3346 * tell the client and log the problem.
3348 if (strlen(csp->headers->first->str) > 30)
3350 csp->headers->first->str[30] = '\0';
3352 log_error(LOG_LEVEL_ERROR,
3353 "Invalid server or forwarder response. Starts with: %s",
3354 csp->headers->first->str);
3355 log_error(LOG_LEVEL_CLF,
3356 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3357 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3358 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3359 free_http_request(http);
3360 mark_server_socket_tainted(csp);
3365 * We have now received the entire server header,
3366 * filter it and send the result to the client
3368 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
3370 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
3372 hdr = list_to_text(csp->headers);
3375 /* FIXME Should handle error properly */
3376 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3379 if (crunch_response_triggered(csp, crunchers_light))
3382 * One of the tags created by a server-header
3383 * tagger triggered a crunch. We already
3384 * delivered the crunch response to the client
3385 * and are done here after cleaning up.
3388 mark_server_socket_tainted(csp);
3391 /* Buffer and pcrs filter this if appropriate. */
3393 if (!http->ssl) /* We talk plaintext */
3395 content_filter = get_filter_function(csp);
3398 * Only write if we're not buffering for content modification
3400 if (!content_filter)
3403 * Write the server's (modified) header to
3404 * the client (along with anything else that
3405 * may be in the buffer)
3408 if (write_socket(csp->cfd, hdr, strlen(hdr))
3409 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
3411 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
3414 * The write failed, so don't bother mentioning it
3415 * to the client... it probably can't hear us anyway.
3418 mark_server_socket_tainted(csp);
3422 byte_count += (unsigned long long)len;
3427 * XXX: the header lenght should probably
3428 * be calculated by get_server_headers().
3430 long header_length = csp->iob->cur - header_start;
3431 assert(csp->iob->cur > header_start);
3432 byte_count += (unsigned long long)(len - header_length);
3435 /* we're finished with the server's header */
3441 * If this was a MS IIS/5 hack then it means the server
3442 * has already closed the connection. Nothing more to read.
3447 log_error(LOG_LEVEL_ERROR,
3448 "Closed server connection detected. "
3449 "Applying the MS IIS5 hack didn't help.");
3450 log_error(LOG_LEVEL_CLF,
3451 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3452 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3453 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3454 mark_server_socket_tainted(csp);
3460 mark_server_socket_tainted(csp);
3461 return; /* huh? we should never get here */
3464 if (csp->content_length == 0)
3467 * If Privoxy didn't recalculate the Content-Lenght,
3468 * byte_count is still correct.
3470 csp->content_length = byte_count;
3473 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3474 if ((csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
3475 && (csp->expected_content_length != byte_count))
3477 log_error(LOG_LEVEL_CONNECT,
3478 "Received %llu bytes while expecting %llu.",
3479 byte_count, csp->expected_content_length);
3480 mark_server_socket_tainted(csp);
3484 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %llu",
3485 csp->ip_addr_str, http->ocmd, csp->content_length);
3489 /*********************************************************************
3493 * Description : This is little more than chat. We only "serve" to
3494 * to close (or remember) any socket that chat may have
3498 * 1 : csp = Current client state (buffers, headers, etc...)
3502 *********************************************************************/
3504 void serve(struct client_state *csp)
3505 #else /* ifndef AMIGA */
3506 static void serve(struct client_state *csp)
3507 #endif /* def AMIGA */
3509 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3510 int continue_chatting = 0;
3515 continue_chatting = (csp->config->feature_flags
3516 & RUNTIME_FEATURE_CONNECTION_KEEP_ALIVE)
3517 && (csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE)
3518 && (csp->cfd != JB_INVALID_SOCKET)
3519 && (csp->sfd != JB_INVALID_SOCKET)
3520 && socket_is_still_usable(csp->sfd);
3523 * Get the csp in a mostly vergin state again.
3524 * XXX: Should be done elsewhere.
3526 csp->content_type = 0;
3527 csp->content_length = 0;
3528 csp->expected_content_length = 0;
3529 list_remove_all(csp->headers);
3530 freez(csp->iob->buf);
3531 memset(csp->iob, 0, sizeof(csp->iob));
3532 freez(csp->error_message);
3533 free_http_request(csp->http);
3534 destroy_list(csp->headers);
3535 destroy_list(csp->tags);
3536 free_current_action(csp->action);
3537 if (NULL != csp->fwd)
3539 unload_forward_spec(csp->fwd);
3543 /* XXX: Store per-connection flags someplace else. */
3544 csp->flags = CSP_FLAG_ACTIVE | (csp->flags & CSP_FLAG_TOGGLED_ON);
3546 if (continue_chatting)
3548 log_error(LOG_LEVEL_CONNECT,
3549 "Waiting for the next client request. "
3550 "Keeping the server socket %d to %s open.",
3551 csp->sfd, csp->server_connection.host);
3553 if ((csp->flags & CSP_FLAG_CLIENT_CONNECTION_KEEP_ALIVE)
3554 && data_is_available(csp->cfd, csp->config->keep_alive_timeout)
3555 && socket_is_still_usable(csp->cfd))
3557 log_error(LOG_LEVEL_CONNECT, "Client request arrived in "
3558 "time or the client closed the connection.");
3562 log_error(LOG_LEVEL_CONNECT,
3563 "No additional client request received in time. "
3564 "Closing server socket %d, initially opened for %s.",
3565 csp->sfd, csp->server_connection.host);
3569 else if (csp->sfd != JB_INVALID_SOCKET)
3571 log_error(LOG_LEVEL_CONNECT,
3572 "The connection on server socket %d to %s isn't reusable. "
3573 "Closing.", csp->sfd, csp->server_connection.host);
3575 } while (continue_chatting);
3578 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3580 if (csp->sfd != JB_INVALID_SOCKET)
3582 close_socket(csp->sfd);
3585 if (csp->cfd != JB_INVALID_SOCKET)
3587 close_socket(csp->cfd);
3590 csp->flags &= ~CSP_FLAG_ACTIVE;
3596 /*********************************************************************
3598 * Function : server_thread
3600 * Description : We only exist to call `serve' in a threaded environment.
3603 * 1 : data = Current client state (buffers, headers, etc...)
3605 * Returns : Always 0.
3607 *********************************************************************/
3608 static int32 server_thread(void *data)
3610 serve((struct client_state *) data);
3617 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3618 /*********************************************************************
3622 * Description : Print usage info & exit.
3624 * Parameters : Pointer to argv[0] for identifying ourselves
3628 *********************************************************************/
3629 static void usage(const char *myname)
3631 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3635 #endif /* defined(unix) */
3638 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3639 #endif /* defined(unix) */
3640 "[--version] [configfile]\n"
3641 "Aborting\n", myname);
3646 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3649 #ifdef MUTEX_LOCKS_AVAILABLE
3650 /*********************************************************************
3652 * Function : privoxy_mutex_lock
3654 * Description : Locks a mutex.
3657 * 1 : mutex = The mutex to lock.
3659 * Returns : Void. May exit in case of errors.
3661 *********************************************************************/
3662 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3664 #ifdef FEATURE_PTHREAD
3665 int err = pthread_mutex_lock(mutex);
3668 if (mutex != &log_mutex)
3670 log_error(LOG_LEVEL_FATAL,
3671 "Mutex locking failed: %s.\n", strerror(err));
3676 EnterCriticalSection(mutex);
3677 #endif /* def FEATURE_PTHREAD */
3681 /*********************************************************************
3683 * Function : privoxy_mutex_unlock
3685 * Description : Unlocks a mutex.
3688 * 1 : mutex = The mutex to unlock.
3690 * Returns : Void. May exit in case of errors.
3692 *********************************************************************/
3693 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3695 #ifdef FEATURE_PTHREAD
3696 int err = pthread_mutex_unlock(mutex);
3699 if (mutex != &log_mutex)
3701 log_error(LOG_LEVEL_FATAL,
3702 "Mutex unlocking failed: %s.\n", strerror(err));
3707 LeaveCriticalSection(mutex);
3708 #endif /* def FEATURE_PTHREAD */
3712 /*********************************************************************
3714 * Function : privoxy_mutex_init
3716 * Description : Prepares a mutex.
3719 * 1 : mutex = The mutex to initialize.
3721 * Returns : Void. May exit in case of errors.
3723 *********************************************************************/
3724 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3726 #ifdef FEATURE_PTHREAD
3727 int err = pthread_mutex_init(mutex, 0);
3730 printf("Fatal error. Mutex initialization failed: %s.\n",
3735 InitializeCriticalSection(mutex);
3736 #endif /* def FEATURE_PTHREAD */
3738 #endif /* def MUTEX_LOCKS_AVAILABLE */
3740 /*********************************************************************
3742 * Function : initialize_mutexes
3744 * Description : Prepares mutexes if mutex support is available.
3748 * Returns : Void, exits in case of errors.
3750 *********************************************************************/
3751 static void initialize_mutexes(void)
3753 #ifdef MUTEX_LOCKS_AVAILABLE
3755 * Prepare global mutex semaphores
3757 privoxy_mutex_init(&log_mutex);
3758 privoxy_mutex_init(&log_init_mutex);
3759 privoxy_mutex_init(&connection_reuse_mutex);
3762 * XXX: The assumptions below are a bit naive
3763 * and can cause locks that aren't necessary.
3765 * For example older FreeBSD versions (< 6.x?)
3766 * have no gethostbyname_r, but gethostbyname is
3769 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3770 privoxy_mutex_init(&resolver_mutex);
3771 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3773 * XXX: should we use a single mutex for
3774 * localtime() and gmtime() as well?
3776 #ifndef HAVE_GMTIME_R
3777 privoxy_mutex_init(&gmtime_mutex);
3778 #endif /* ndef HAVE_GMTIME_R */
3780 #ifndef HAVE_LOCALTIME_R
3781 privoxy_mutex_init(&localtime_mutex);
3782 #endif /* ndef HAVE_GMTIME_R */
3785 privoxy_mutex_init(&rand_mutex);
3786 #endif /* ndef HAVE_RANDOM */
3787 #endif /* def MUTEX_LOCKS_AVAILABLE */
3791 /*********************************************************************
3795 * Description : Load the config file and start the listen loop.
3796 * This function is a lot more *sane* with the `load_config'
3797 * and `listen_loop' functions; although it stills does
3798 * a *little* too much for my taste.
3801 * 1 : argc = Number of parameters (including $0).
3802 * 2 : argv = Array of (char *)'s to the parameters.
3804 * Returns : 1 if : can't open config file, unrecognized directive,
3805 * stats requested in multi-thread mode, can't open the
3806 * log file, can't open the jar file, listen port is invalid,
3807 * any load fails, and can't bind port.
3809 * Else main never returns, the process must be signaled
3810 * to terminate execution. Or, on Windows, use the
3811 * "File", "Exit" menu option.
3813 *********************************************************************/
3815 int real_main(int argc, const char *argv[])
3817 int main(int argc, const char *argv[])
3821 unsigned int random_seed;
3823 struct passwd *pw = NULL;
3824 struct group *grp = NULL;
3827 char *pre_chroot_nslookup_to_load_resolver = NULL;
3834 #if !defined(_WIN32)
3841 /* Prepare mutexes if supported and necessary. */
3842 initialize_mutexes();
3844 /* Enable logging until further notice. */
3848 * Parse the command line arguments
3850 * XXX: simply printing usage information in case of
3851 * invalid arguments isn't particularly user friendly.
3853 while (++argc_pos < argc)
3856 /* Check to see if the service must be installed or uninstalled */
3857 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3859 const char *pName = argv[argc_pos] + 9;
3862 exit( (install_service(pName)) ? 0 : 1 );
3864 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3866 const char *pName = argv[argc_pos] + 11;
3869 exit((uninstall_service(pName)) ? 0 : 1);
3871 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3873 bRunAsService = TRUE;
3874 w32_set_service_cwd();
3875 atexit(w32_service_exit_notify);
3878 #endif /* defined(_WIN32) */
3881 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3883 if (strcmp(argv[argc_pos], "--help") == 0)
3888 else if(strcmp(argv[argc_pos], "--version") == 0)
3890 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3896 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3898 set_debug_level(LOG_LEVEL_FATAL | LOG_LEVEL_ERROR | LOG_LEVEL_INFO);
3902 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3904 if (++argc_pos == argc) usage(argv[0]);
3905 pidfile = strdup(argv[argc_pos]);
3908 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3910 if (++argc_pos == argc) usage(argv[argc_pos]);
3912 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3915 if (NULL == (grp = getgrnam(p)))
3917 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3921 if (NULL == (pw = getpwnam(argv[argc_pos])))
3923 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3926 if (p != NULL) *--p = '\0';
3929 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3931 if (++argc_pos == argc) usage(argv[0]);
3932 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3935 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3939 #endif /* defined(unix) */
3941 else if (argc_pos + 1 != argc)
3944 * This is neither the last command line
3945 * option, nor was it recognized before,
3946 * therefore it must be invalid.
3952 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3954 configfile = argv[argc_pos];
3957 } /* -END- while (more arguments) */
3959 show_version(Argv[0]);
3962 if ( *configfile != '/' )
3964 char cwd[BUFFER_SIZE];
3966 size_t abs_file_size;
3968 /* make config-filename absolute here */
3969 if (NULL == getcwd(cwd, sizeof(cwd)))
3971 perror("failed to get current working directory");
3976 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3977 basedir = strdup(cwd);
3979 if (NULL == basedir ||
3980 NULL == (abs_file = malloc(abs_file_size)))
3982 perror("malloc failed");
3985 strlcpy(abs_file, basedir, abs_file_size);
3986 strlcat(abs_file, "/", abs_file_size );
3987 strlcat(abs_file, configfile, abs_file_size);
3988 configfile = abs_file;
3990 #endif /* defined unix */
3994 clients->next = NULL;
3996 /* XXX: factor out initialising after the next stable release. */
3999 #elif defined(_WIN32)
4003 random_seed = (unsigned int)time(NULL);
4005 srandom(random_seed);
4008 #endif /* ifdef HAVE_RANDOM */
4011 * Unix signal handling
4013 * Catch the abort, interrupt and terminate signals for a graceful exit
4014 * Catch the hangup signal so the errlog can be reopened.
4015 * Ignore the broken pipe signals (FIXME: Why?)
4017 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
4020 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
4021 const int ignored_signals[] = { SIGPIPE, 0 };
4023 for (idx = 0; catched_signals[idx] != 0; idx++)
4025 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
4026 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
4028 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
4029 #endif /* ifdef sun */
4031 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
4035 for (idx = 0; ignored_signals[idx] != 0; idx++)
4037 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
4039 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
4044 #else /* ifdef _WIN32 */
4045 # ifdef _WIN_CONSOLE
4047 * We *are* in a windows console app.
4048 * Print a verbose messages about FAQ's and such
4050 printf("%s", win32_blurb);
4051 # endif /* def _WIN_CONSOLE */
4052 #endif /* def _WIN32 */
4055 /* Initialize the CGI subsystem */
4056 cgi_init_error_messages();
4059 * If runnig on unix and without the --nodaemon
4060 * option, become a daemon. I.e. fork, detach
4061 * from tty and get process group leadership
4074 if ( pid < 0 ) /* error */
4079 else if ( pid != 0 ) /* parent */
4084 * must check for errors
4085 * child died due to missing files aso
4088 wpid = waitpid( pid, &status, WNOHANG );
4097 /* Should be more portable, but not as well tested */
4102 #else /* ndef __FreeBSD__ */
4104 #endif /* ndef __FreeBSD__ */
4105 fd = open("/dev/tty", O_RDONLY);
4108 /* no error check here */
4109 ioctl( fd, TIOCNOTTY,0 );
4114 * stderr (fd 2) will be closed later on,
4115 * when the config file has been parsed.
4122 } /* -END- if (!no_daemon) */
4125 * As soon as we have written the PID file, we can switch
4126 * to the user and group ID indicated by the --user option
4132 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
4134 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
4138 if (setgroups(1, &grp->gr_gid))
4140 log_error(LOG_LEVEL_FATAL, "setgroups() failed: %E");
4143 else if (initgroups(pw->pw_name, pw->pw_gid))
4145 log_error(LOG_LEVEL_FATAL, "initgroups() failed: %E");
4151 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
4153 /* Read the time zone file from /etc before doing chroot. */
4155 if (NULL != pre_chroot_nslookup_to_load_resolver
4156 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
4158 /* Initialize resolver library. */
4159 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
4161 if (chroot(pw->pw_dir) < 0)
4163 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
4167 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
4170 if (setuid(pw->pw_uid))
4172 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
4176 char putenv_dummy[64];
4178 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
4179 if (putenv(putenv_dummy) != 0)
4181 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
4184 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
4185 if (putenv(putenv_dummy) != 0)
4187 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
4193 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
4196 #endif /* defined unix */
4199 /* This will be FALSE unless the command line specified --service
4203 /* Yup, so now we must attempt to establish a connection
4204 * with the service dispatcher. This will only work if this
4205 * process was launched by the service control manager to
4206 * actually run as a service. If this isn't the case, i've
4207 * known it take around 30 seconds or so for the call to return.
4210 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
4211 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
4213 /* Service has run, and at this point is now being stopped, so just return */
4218 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
4220 /* An error occurred. Usually it's because --service was wrongly specified
4221 * and we were unable to connect to the Service Control Dispatcher because
4222 * it wasn't expecting us and is therefore not listening.
4224 * For now, just continue below to call the listen_loop function.
4227 #endif /* def _WIN32 */
4237 /*********************************************************************
4239 * Function : bind_port_helper
4241 * Description : Bind the listen port. Handles logging, and aborts
4245 * 1 : config = Privoxy configuration. Specifies port
4248 * Returns : Port that was opened.
4250 *********************************************************************/
4251 static jb_socket bind_port_helper(struct configuration_spec * config)
4256 if (config->haddr == NULL)
4258 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
4263 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
4264 config->hport, config->haddr);
4267 result = bind_port(config->haddr, config->hport, &bfd);
4274 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
4275 "There may be another Privoxy or some other "
4276 "proxy running on port %d",
4277 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
4278 config->hport, config->hport);
4281 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
4282 "The hostname is not resolvable",
4283 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
4286 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
4287 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
4290 /* shouldn't get here */
4291 return JB_INVALID_SOCKET;
4294 config->need_bind = 0;
4301 /* Without this simple workaround we get this compiler warning from _beginthread
4302 * warning C4028: formal parameter 1 different from declaration
4304 void w32_service_listen_loop(void *p)
4308 #endif /* def _WIN32 */
4311 /*********************************************************************
4313 * Function : listen_loop
4315 * Description : bind the listen port and enter a "FOREVER" listening loop.
4321 *********************************************************************/
4322 static void listen_loop(void)
4324 struct client_state *csp = NULL;
4326 struct configuration_spec *config;
4327 unsigned int active_threads = 0;
4329 config = load_config();
4331 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
4333 * XXX: Should be relocated once it no
4334 * longer needs to emit log messages.
4336 initialize_reusable_connections();
4337 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
4339 bfd = bind_port_helper(config);
4341 #ifdef FEATURE_GRACEFUL_TERMINATION
4342 while (!g_terminate)
4347 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
4348 while (waitpid(-1, NULL, WNOHANG) > 0)
4350 /* zombie children */
4352 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
4355 * Free data that was used by died threads
4357 active_threads = sweep();
4361 * Re-open the errlog after HUP signal
4363 if (received_hup_signal)
4365 if (NULL != config->logfile)
4367 init_error_log(Argv[0], config->logfile);
4369 received_hup_signal = 0;
4373 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
4375 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
4379 csp->flags |= CSP_FLAG_ACTIVE;
4380 csp->sfd = JB_INVALID_SOCKET;
4382 csp->config = config = load_config();
4384 if ( config->need_bind )
4387 * Since we were listening to the "old port", we will not see
4388 * a "listen" param change until the next request. So, at
4389 * least 1 more request must be made for us to find the new
4390 * setting. I am simply closing the old socket and binding the
4393 * Which-ever is correct, we will serve 1 more page via the
4394 * old settings. This should probably be a "show-proxy-args"
4395 * request. This should not be a so common of an operation
4396 * that this will hurt people's feelings.
4401 bfd = bind_port_helper(config);
4404 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
4406 if (!accept_connection(csp, bfd))
4408 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
4421 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
4424 #ifdef FEATURE_TOGGLE
4425 if (global_toggle_state)
4426 #endif /* def FEATURE_TOGGLE */
4428 csp->flags |= CSP_FLAG_TOGGLED_ON;
4431 if (run_loader(csp))
4433 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
4434 /* Never get here - LOG_LEVEL_FATAL causes program exit */
4438 if (block_acl(NULL,csp))
4440 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
4441 close_socket(csp->cfd);
4442 freez(csp->ip_addr_str);
4446 #endif /* def FEATURE_ACL */
4448 if ((0 != config->max_client_connections)
4449 && (active_threads >= config->max_client_connections))
4451 log_error(LOG_LEVEL_CONNECT,
4452 "Rejecting connection from %s. Maximum number of connections reached.",
4454 write_socket(csp->cfd, TOO_MANY_CONNECTIONS_RESPONSE,
4455 strlen(TOO_MANY_CONNECTIONS_RESPONSE));
4456 close_socket(csp->cfd);
4457 freez(csp->ip_addr_str);
4462 /* add it to the list of clients */
4463 csp->next = clients->next;
4464 clients->next = csp;
4466 if (config->multi_threaded)
4470 /* this is a switch () statment in the C preprocessor - ugh */
4471 #undef SELECTED_ONE_OPTION
4473 /* Use Pthreads in preference to native code */
4474 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
4475 #define SELECTED_ONE_OPTION
4477 pthread_t the_thread;
4478 pthread_attr_t attrs;
4480 pthread_attr_init(&attrs);
4481 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
4482 errno = pthread_create(&the_thread, &attrs,
4483 (void * (*)(void *))serve, csp);
4484 child_id = errno ? -1 : 0;
4485 pthread_attr_destroy(&attrs);
4489 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
4490 #define SELECTED_ONE_OPTION
4491 child_id = _beginthread(
4492 (void (*)(void *))serve,
4497 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
4498 #define SELECTED_ONE_OPTION
4499 child_id = _beginthread(
4500 (void(* _Optlink)(void*))serve,
4506 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
4507 #define SELECTED_ONE_OPTION
4509 thread_id tid = spawn_thread
4510 (server_thread, "server", B_NORMAL_PRIORITY, csp);
4512 if ((tid >= 0) && (resume_thread(tid) == B_OK))
4514 child_id = (int) tid;
4523 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
4524 #define SELECTED_ONE_OPTION
4525 csp->cfd = ReleaseSocket(csp->cfd, -1);
4528 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
4529 NP_Output, Output(),
4530 NP_CloseOutput, FALSE,
4531 NP_Name, (ULONG)"privoxy child",
4535 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
4536 NP_Output, Output(),
4537 NP_CloseOutput, FALSE,
4538 NP_Name, (ULONG)"privoxy child",
4539 NP_StackSize, 200*1024,
4545 ((struct Task *)child_id)->tc_UserData = csp;
4546 Signal((struct Task *)child_id, SIGF_SINGLE);
4551 #if !defined(SELECTED_ONE_OPTION)
4554 /* This block is only needed when using fork().
4555 * When using threads, the server thread was
4556 * created and run by the call to _beginthread().
4558 if (child_id == 0) /* child */
4561 #ifdef FEATURE_TOGGLE
4562 int inherited_toggle_state = global_toggle_state;
4563 #endif /* def FEATURE_TOGGLE */
4568 * If we've been toggled or we've blocked the request, tell Mom
4571 #ifdef FEATURE_TOGGLE
4572 if (inherited_toggle_state != global_toggle_state)
4574 rc |= RC_FLAG_TOGGLED;
4576 #endif /* def FEATURE_TOGGLE */
4578 #ifdef FEATURE_STATISTICS
4579 if (csp->flags & CSP_FLAG_REJECTED)
4581 rc |= RC_FLAG_BLOCKED;
4583 #endif /* ndef FEATURE_STATISTICS */
4587 else if (child_id > 0) /* parent */
4589 /* in a fork()'d environment, the parent's
4590 * copy of the client socket and the CSP
4594 #if !defined(_WIN32) && !defined(__CYGWIN__)
4596 wait( &child_status );
4599 * Evaluate child's return code: If the child has
4600 * - been toggled, toggle ourselves
4601 * - blocked its request, bump up the stats counter
4604 #ifdef FEATURE_TOGGLE
4605 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
4607 global_toggle_state = !global_toggle_state;
4609 #endif /* def FEATURE_TOGGLE */
4611 #ifdef FEATURE_STATISTICS
4613 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4617 #endif /* def FEATURE_STATISTICS */
4619 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4620 close_socket(csp->cfd);
4621 csp->flags &= ~CSP_FLAG_ACTIVE;
4625 #undef SELECTED_ONE_OPTION
4626 /* end of cpp switch () */
4631 * Spawning the child failed, assume it's because
4632 * there are too many children running already.
4633 * XXX: If you assume ...
4635 log_error(LOG_LEVEL_ERROR,
4636 "Unable to take any additional connections: %E");
4637 write_socket(csp->cfd, TOO_MANY_CONNECTIONS_RESPONSE,
4638 strlen(TOO_MANY_CONNECTIONS_RESPONSE));
4639 close_socket(csp->cfd);
4640 csp->flags &= ~CSP_FLAG_ACTIVE;
4649 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4651 /* Clean up. Aim: free all memory (no leaks) */
4652 #ifdef FEATURE_GRACEFUL_TERMINATION
4654 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4656 unload_current_config_file();
4657 unload_current_actions_file();
4658 unload_current_re_filterfile();
4659 #ifdef FEATURE_TRUST
4660 unload_current_trust_file();
4663 if (config->multi_threaded)
4670 } while ((clients->next != NULL) && (--i > 0));
4674 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4685 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4686 /* Cleanup - remove taskbar icon etc. */
4691 #endif /* FEATURE_GRACEFUL_TERMINATION */