1 const char jcc_rcs[] = "$Id: jcc.c,v 1.246 2009/05/10 10:12:30 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 1.246 2009/05/10 10:12:30 fabiankeil
37 * Initial keep-alive support for the client socket.
38 * Temporarily disable the server-side-only keep-alive code.
40 * Revision 1.245 2009/04/24 15:29:43 fabiankeil
41 * Allow to limit the number of of client connections.
43 * Revision 1.244 2009/04/17 11:34:34 fabiankeil
44 * Style cosmetics for the IPv6 code.
46 * Revision 1.243 2009/04/17 11:27:49 fabiankeil
47 * Petr Pisar's privoxy-3.0.12-ipv6-3.diff.
49 * Revision 1.242 2009/04/11 10:44:47 fabiankeil
50 * Update a comment. We're not in Kansas anymore.
52 * Revision 1.241 2009/04/11 10:37:23 fabiankeil
53 * When dropping connections due to ACL, don't leak csp->ip_addr_str.
55 * Revision 1.240 2009/04/09 10:12:54 fabiankeil
56 * Fix two cases in which an invalid server response would result
57 * in the client connection being closed without sending an error
60 * Revision 1.239 2009/04/07 11:43:50 fabiankeil
61 * If the server rudely resets the connection directly after sending the
62 * headers, pass the mess to the client instead of sending an incorrect
63 * connect-failed message. Fixes #2698674 reported by mybugaccount.
65 * Revision 1.238 2009/03/27 14:42:30 fabiankeil
66 * Correct the status code for CONNECTION_TIMEOUT_RESPONSE.
68 * Revision 1.237 2009/03/27 14:32:04 fabiankeil
69 * If spawning a child in listen_loop() fails, send a real
70 * HTTP response to the client and continue listening for
71 * new connections without artificial delay.
73 * Revision 1.236 2009/03/25 17:30:24 fabiankeil
74 * In serve(), keep the client socket open until we marked the
75 * server socket as unused. This should increase the chances
76 * that we reuse the connection for the client's next request
77 * to the same destination.
79 * Revision 1.235 2009/03/18 21:01:20 fabiankeil
80 * Comment fix. Spotted by Roland.
82 * Revision 1.234 2009/03/18 20:48:42 fabiankeil
83 * If the --no-daemon option is used, enable LOG_LEVEL_INFO
84 * before the config file has been parsed (as we always did).
86 * Revision 1.233 2009/03/13 14:10:07 fabiankeil
87 * Fix some more harmless warnings on amd64.
89 * Revision 1.232 2009/03/08 19:29:16 fabiankeil
90 * Reinitialize the timeout structure every time before passing
91 * it to select(). Apparently some implementations mess with it.
92 * Probably fixes #2669131 reported by cyberpatrol.
94 * Revision 1.231 2009/03/08 14:19:23 fabiankeil
95 * Fix justified (but harmless) compiler warnings
96 * on platforms where sizeof(int) < sizeof(long).
98 * Revision 1.230 2009/03/07 13:09:17 fabiankeil
99 * Change csp->expected_content and_csp->expected_content_length from
100 * size_t to unsigned long long to reduce the likelihood of integer
101 * overflows that would let us close the connection prematurely.
102 * Bug found while investigating #2669131, reported by cyberpatrol.
104 * Revision 1.229 2009/03/07 11:17:01 fabiankeil
105 * Fix compiler warning.
107 * Revision 1.228 2009/03/06 20:30:13 fabiankeil
108 * Log unsigned values as such.
110 * Revision 1.227 2009/03/02 19:18:11 fabiankeil
111 * Streamline parse_http_request()'s prototype. As
112 * cparser pointed out it doesn't actually use csp.
114 * Revision 1.226 2009/03/01 18:28:24 fabiankeil
115 * Help clang understand that we aren't dereferencing
116 * NULL pointers here.
118 * Revision 1.225 2009/02/19 18:09:32 fabiankeil
119 * Unbreak build without FEATURE_CONNECTION_KEEP_ALIVE.
122 * Revision 1.224 2009/02/14 15:32:04 fabiankeil
123 * Add the request URL to the timeout message in chat().
126 * Revision 1.223 2009/02/09 21:21:16 fabiankeil
127 * Now that init_log_module() is called earlier, call show_version()
128 * later on from main() directly so it doesn't get called for --help
131 * Revision 1.222 2009/02/08 12:56:51 fabiankeil
132 * Call initialize_mutexes() before init_log_module() again.
133 * Broken since r220, might be the cause of Lee's #2579448.
135 * Revision 1.221 2009/02/06 18:02:58 fabiankeil
136 * When dropping privileges, also give up membership in supplementary
137 * groups. Thanks to Matthias Drochner for reporting the problem,
138 * providing the initial patch and testing the final version.
140 * Revision 1.220 2009/02/04 18:29:07 fabiankeil
141 * Initialize the log module before parsing arguments.
142 * Thanks to Matthias Drochner for the report.
144 * Revision 1.219 2009/01/31 16:08:21 fabiankeil
145 * Remove redundant error check in receive_client_request().
147 * Revision 1.218 2009/01/31 12:25:54 fabiankeil
148 * Flatten indentation in receive_client_request().
150 * Revision 1.217 2009/01/07 19:50:09 fabiankeil
151 * - If the socket-timeout has been reached and the client
152 * hasn't received any data yet, send an explanation before
153 * closing the connection.
154 * - In get_request_line(), signal timeouts the right way.
156 * Revision 1.216 2008/12/24 22:13:11 ler762
157 * fix GCC 3.4.4 warning
159 * Revision 1.215 2008/12/24 17:06:19 fabiankeil
160 * Keep a thread around to timeout alive connections
161 * even if no new requests are coming in.
163 * Revision 1.214 2008/12/20 14:53:55 fabiankeil
164 * Add config option socket-timeout to control the time
165 * Privoxy waits for data to arrive on a socket. Useful
166 * in case of stale ssh tunnels or when fuzz-testing.
168 * Revision 1.213 2008/12/15 18:45:51 fabiankeil
169 * When logging crunches, log the whole URL, so one can easily
170 * differentiate between vanilla HTTP and CONNECT requests.
172 * Revision 1.212 2008/12/14 15:46:22 fabiankeil
173 * Give crunched requests their own log level.
175 * Revision 1.211 2008/12/06 10:05:03 fabiankeil
176 * Downgrade "Received x bytes while expecting y." message to
177 * LOG_LEVEL_CONNECT as it doesn't necessarily indicate an error.
179 * Revision 1.210 2008/12/02 22:03:18 fabiankeil
180 * Don't miscalculate byte_count if we don't get all the
181 * server headers with one read_socket() call. With keep-alive
182 * support enabled, this caused delays until the server closed
185 * Revision 1.209 2008/11/27 09:44:04 fabiankeil
186 * Cosmetics for the last commit: Don't watch out for
187 * the last chunk if the content isn't chunk-encoded or
188 * if we already determined the content length previously.
190 * Revision 1.208 2008/11/26 18:24:17 fabiankeil
191 * Recognize that the server response is complete if the
192 * last chunk is read together with the server headers.
195 * Revision 1.207 2008/11/25 17:25:16 fabiankeil
196 * Don't convert the client-header list to text until we need to.
198 * Revision 1.206 2008/11/23 17:00:11 fabiankeil
199 * Some more chat() cosmetics.
201 * Revision 1.205 2008/11/16 12:43:49 fabiankeil
202 * Turn keep-alive support into a runtime feature
203 * that is disabled by setting keep-alive-timeout
204 * to a negative value.
206 * Revision 1.204 2008/11/06 19:42:17 fabiankeil
207 * Fix last-chunk detection hack to also apply
208 * if buf[] contains nothing but the last-chunk.
210 * Revision 1.203 2008/11/06 18:34:35 fabiankeil
211 * Factor receive_client_request() and
212 * parse_client_request() out of chat().
214 * Revision 1.202 2008/11/02 18:40:34 fabiankeil
215 * If we received a different amount of data than we expected,
216 * log a warning and make sure the server socket isn't reused.
218 * Revision 1.201 2008/11/02 16:48:20 fabiankeil
219 * Revert revision 1.195 and try again.
221 * Revision 1.200 2008/10/26 16:53:18 fabiankeil
224 * Revision 1.199 2008/10/26 15:36:10 fabiankeil
225 * Remove two debug messages with LOG_LEVEL_INFO.
227 * Revision 1.198 2008/10/22 15:19:55 fabiankeil
228 * Once More, With Feeling: if there is no logfile
229 * because the user didn't specify one, we shouldn't
230 * call init_error_log() after receiving SIGHUP either.
232 * Revision 1.197 2008/10/20 17:02:40 fabiankeil
233 * If SIGHUP is received while we aren't running in daemon
234 * mode, calling init_error_log() would be a mistake.
236 * Revision 1.196 2008/10/16 09:16:41 fabiankeil
237 * - Fix two gcc44 conversion warnings.
238 * - Don't bother logging the last five bytes
241 * Revision 1.195 2008/10/13 16:04:37 fabiankeil
242 * Make sure we don't try to reuse tainted server sockets.
244 * Revision 1.194 2008/10/12 18:35:18 fabiankeil
245 * The last commit was a bit too ambitious, apparently the content
246 * length adjustment is only necessary if we aren't buffering.
248 * Revision 1.193 2008/10/12 15:57:35 fabiankeil
249 * Fix content length calculation if we read headers
250 * and the start of the body at once. Now that we have
251 * FEATURE_CONNECTION_KEEP_ALIVE, it actually matters.
253 * Revision 1.192 2008/10/11 18:19:14 fabiankeil
254 * Even more chat() cosmetics.
256 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
257 * Reformat some comments in chat().
259 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
260 * In case of chunk-encoded content, stop reading if
261 * the buffer looks like it ends with the last chunk.
263 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
264 * Let server_response_is_complete() deal properly with
265 * content that is neither buffered nor read all at once.
267 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
268 * Flush work-in-progress changes to keep outgoing connections
269 * alive where possible. Incomplete and mostly #ifdef'd out.
271 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
272 * Add mutex lock support for _WIN32.
274 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
275 * Prepare for critical sections on Windows by adding a
276 * layer of indirection before the pthread mutex functions.
278 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
279 * Remove FEATURE_COOKIE_JAR.
281 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
282 * - Silence LLVM/Clang complaint.
283 * - Make received_hup_signal static.
284 * - Hide definitions for basedir, pidfile and received_hup_signal
285 * from __EMX__ as they only seem to be used in case of #ifdef unix.
287 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
288 * Accept Shoutcast responses again. Problem reported
289 * and fix suggested by Stefan in #2062860.
291 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
292 * Fix possible NULL-pointer dereference reported
293 * by din_a4 in #2003937. Pointy hat to me.
295 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
296 * Streamline sed()'s prototype and declare
297 * the header parse and add structures static.
299 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
300 * - Mark csp as immutable for send_crunch_response().
301 * - Fix comment spelling.
303 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
304 * Factor update_server_headers() out of sed(), ditch the
305 * first_run hack and make server_patterns_light static.
307 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
308 * Don't provide get_header() with the whole client state
309 * structure when it only needs access to csp->iob.
311 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
312 * Make the "read the rest of the headers" loop a bit more readable.
314 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
315 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
316 * - Remove useless comment.
318 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
319 * Fix comment grammar.
321 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
322 * Remove the pointless buffer in client_protocol_is_unsupported().
324 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
325 * Least-effort fix for bug #1821930 (reported by Lee):
326 * If the response doesn't look like HTTP,
327 * tell the client and log the problem.
329 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
330 * Don't pass the whole csp structure to flush_socket()
331 * when it only needs a file descriptor and a buffer.
333 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
334 * Remove kill-popups action.
336 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
337 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
339 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
340 * Remove the treat-forbidden-connects-like-blocks action. We now
341 * use the "blocked" page for forbidden CONNECT requests by default.
343 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
344 * Also use shiny new connect_port_is_forbidden() in jcc.c.
346 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
347 * Rename url_actions() to get_url_actions() and let it
348 * use the standard parameter ordering.
350 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
351 * Let forward_url() use the standard parameter ordering
352 * and mark its second parameter immutable.
354 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
355 * Remove the "Listening ... for local connections only" log message.
356 * Whether or not remote connections are able to reach Privoxy is up
357 * to the operating system.
359 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
360 * Prevent the log messages for CONNECT requests to unacceptable
361 * ports from printing the limit-connect argument as [null] if
362 * limit-connect hasn't been explicitly enabled.
364 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
365 * Make sure all console-mode apps get a usage() instance
367 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
368 * Reword NO_SERVER_DATA_RESPONSE to make it harder
369 * to misunderstand what the message is all about.
371 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
372 * Unbreak trustfile which previously didn't work without
373 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
375 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
376 * Plug memory leak. Spotted by Valgrind, triggered by
377 * Privoxy-Regression-Test feeding proxyfuzz.py.
379 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
380 * In the HTTP snipplets, refer to the client as client.
382 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
383 * Emit a log message when activating the MS IIS5 hack.
385 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
386 * Log the "weak randomization factor" warning only
387 * once for mingw32 and provide some more details.
389 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
390 * Initialize log module after initializing mutexes, future
391 * deadlocks in that code should now work cross-platform.
393 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
394 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
395 * Reported by Bert van Leeuwen in BR#1818808.
397 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
398 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
400 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
401 * When in daemon mode, close stderr after the configuration file has been
402 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
404 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
405 * - Fix a crash when parsing invalid requests whose first header
406 * is rejected by get_header(). Regression (re?)introduced
407 * in r1.143 by yours truly.
408 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
409 * client_cookie_adder() to make sure send-vanilla-wafer can be
410 * controlled through tags (and thus regression-tested).
412 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
413 * - Move get_filter_function() from jcc.c to filters.c
414 * so the filter functions can be static.
415 * - Don't bother filtering body-less responses.
417 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
418 * Execute content filters through execute_content_filter().
420 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
421 * Initialize req to NULL to make sure it's defined if the
422 * first read_socket() call fails. Reported by icmp30.
424 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
425 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
426 * extensive comments moved to user manual.
428 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
429 * Don't crash if a broken header filter wiped out the request line.
431 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
432 * Fix byte_count calculation in case of flushes
433 * and don't parse the server headers a second time.
435 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
436 * - If there's a connection problem after we already forwarded
437 * parts of the original content, just hang up. Fixes BR#1776724.
438 * - Fix warnings about unused code on mingw32.
439 * - In case of flushes, calculate the byte count
440 * less incorrectly (I think).
442 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
443 * Add some more prototypes for static functions.
445 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
446 * Comment out request_contains_null_bytes() until it's used again.
448 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
449 * #1763173 from Stefan Huehner: s@const static@static const@
450 * and declare some more functions static.
452 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
453 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
454 * and explain why they were rejected in the first place.
455 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
456 * CONNECT requests. The request line was missing.
457 * - Add two more XXX reminders as we don't have enough already.
459 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
460 * As Hal noticed, checking dispatch_cgi() as the last cruncher
461 * looks like a bug if CGI requests are blocked unintentionally,
462 * so don't do it unless the user enabled the new config option
463 * "allow-cgi-request-crunching".
465 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
466 * - Allow to rewrite the request destination behind the client's back.
467 * - Turn the weird-looking unconditional for loop that
468 * reads the client request into a conditional while loop.
469 * Move the stuff that only runs once out of the loop.
470 * - Move parts of chat(), server_content_type() and the
471 * necessary stuff to fix BR#1750917 into get_filter_function().
473 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
474 * Temporary workaround for BR#1730105.
476 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
477 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
478 * deadlocks and crashes on OpenBSD and possibly other OS with neither
479 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
480 * Thanks to Ralf Horstmann for report and solution.
482 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
483 * Add forward-override{} to change the forwarding settings through
484 * action sections. This is mainly interesting to forward different
485 * clients differently (for example based on User-Agent or request
488 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
489 * - Let usage() mention the --chroot parameter.
490 * - Use read_socket() consistently and always leave
491 * the last buffer byte alone, even in cases where
492 * null termination (currently) doesn't matter.
494 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
495 * - Fix config file loading on Unix if no config file is specified.
496 * Since r1.97 Privoxy would always interpret the last argument as
497 * config file, even if it's a valid command line option.
498 * - Abort in case of unrecognized command line options. Closes #1719696.
499 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
500 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
502 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
503 * - Don't rerun crunchers that only depend on the request URL.
504 * - Don't count redirects and CGI requests as "blocked requests".
506 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
507 * Support crunching based on tags created by server-header taggers.
509 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
510 * Make HTTP snippets static (again). Add a Content-Type for those
511 * with content so the browser doesn't guess it based on the URL.
513 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
514 * Move crunching and request line rebuilding out of chat().
516 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
517 * Introduce tags as alternative way to specify which
518 * actions apply to a request. At the moment tags can be
519 * created based on client and server headers.
521 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
522 * Don't CLF-log CONNECT requests twice.
524 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
525 * Log the source address for ACL-related connection drops.
527 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
528 * New config option: enforce-blocks.
530 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
531 * - Move null byte check into separate function.
532 * - Don't confuse the client with error pages
533 * if a CONNECT request was already confirmed.
535 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
536 * Speed up NULL byte escaping and only log the complete
537 * NULL byte requests with header debugging enabled.
539 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
540 * Answer requests that contain NULL bytes with
541 * a custom response instead of waiting for more
542 * data until the client eventually hangs up.
544 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
545 * - Move delivery and logging of crunched responses
546 * from chat() into send_crunch_response().
547 * - Display the reason for generating http_responses.
548 * - Log the content length for LOG_LEVEL_CLF correctly
549 * (still incorrect for some fixed responses).
550 * - Reword an incorrect comment about
551 * treat-forbidden-connects-like-blocks violating
553 * - Add some log messages.
555 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
556 * Move mutex initialization into separate
557 * function and exit in case of errors.
559 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
560 * - Start to reduce chat()'s line count and move
561 * parts of it into separate functions.
562 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
564 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
565 * - Add Proxy-Agent header to HTTP snippets that are
566 * supposed to reach HTTP clients only.
567 * - Made a few CONNECT log messages more descriptive.
568 * - Catch completely empty server responses (as seen
569 * with Tor's fake ".noconnect" top level domain).
570 * - Use shiny new "forwarding-failed" template for socks errors.
572 * Revision 1.118 2007/01/07 07:43:43 joergs
573 * AmigaOS4 support added.
575 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
576 * Added config option accept-intercepted-requests
577 * and disabled it by default.
579 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
580 * Reverted parts of my last commit
581 * to keep error handling working.
583 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
584 * Fixed gcc43 conversion warnings.
586 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
587 * Fix -pedantic ISO C warning about converting
588 * from function pointer to object pointer.
590 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
591 * Silence compiler warning I introduced with my last commit.
593 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
594 * Mutex protect rand() if POSIX threading
595 * is used, warn the user if that's not possible
596 * and stop using it on _WIN32 where it could
599 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
600 * Don't prevent core dumps by catching SIGABRT.
601 * It's rude and makes debugging unreasonable painful.
603 * Revision 1.110 2006/12/13 14:52:53 etresoft
604 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
606 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
607 * Privoxy is now able to run as intercepting
608 * proxy in combination with any packet filter
609 * that does the port redirection. The destination
610 * is extracted from the "Host:" header which
611 * should be available for nearly all requests.
613 * Moved HTTP snipplets into jcc.c.
614 * Added error message for gopher proxy requests.
616 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
617 * Only unlink the pidfile if it's actually used.
619 * Change order of interception checks to make
620 * it possible to block or redirect requests for
623 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
624 * Make pthread mutex locking more generic. Instead of
625 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
626 * and use mutex locking unless there is an _r function
627 * available. Better safe than sorry.
629 * Fixes "./configure --disable-pthread" and should result
630 * in less threading-related problems on pthread-using platforms,
631 * but it still doesn't fix BR#1122404.
633 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
634 * Move pthread.h inclusion from jcc.c to jcc.h.
635 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
637 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
638 * Don't exit after receiving the second SIGHUP on Solaris.
640 * Fixes BR 1052235, but the same problem may exist on other
641 * systems. Once 3.0.6 is out we should use sigset()
642 * where available and see if it breaks anything.
644 * Revision 1.104 2006/09/23 13:26:38 roro
645 * Replace TABs by spaces in source code.
647 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
648 * Fix +redirect{}. Didn't work with -fast-redirects.
650 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
651 * Respond with 400 and a short text message
652 * if the client tries to use Privoxy as FTP proxy.
654 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
655 * Make number of retries in case of forwarded-connect problems
656 * a config file option (forwarded-connect-retries) and use 0 as
659 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
660 * Set random(3) seed.
662 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
663 * Follow the OpenBSD port's lead and protect the resolve
664 * functions on OpenBSD as well.
666 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
667 * --user fix. Only use the user as group if no group is specified.
668 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
670 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
671 * Windows service (re-)integration
675 * --install[:service_name]
676 * --uninstall[:service_name]
679 * They work as follows:
680 * --install will create a service for you and then terminate.
681 * By default the service name will be "privoxy" (without the quotes).
682 * However you can run multiple services if you wish, just by adding
683 * a colon and then a name (no spaces).
685 * --uninstall follows the exact same rules a --install.
687 * --service is used when the program is executed by the service
688 * control manager, and in normal circumstances would never be
689 * used as a command line argument.
691 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
692 * Windows service integration
694 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
695 * Incorporate Fabian Keil's patch work:
696 http://www.fabiankeil.de/sourcecode/privoxy/
698 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
699 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
700 * with what was really the latest development (the v_3_0_branch branch)
702 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
703 * Thanks to Jindrich Makovicka for a race condition fix for the log
704 * file. The race condition remains for non-pthread implementations.
705 * Reference patch #1175720.
707 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
708 * Add the ability to check jpeg images for invalid
709 * lengths of comment blocks. Defensive strategy
710 * against the exploit:
711 * Microsoft Security Bulletin MS04-028
712 * Buffer Overrun in JPEG Processing (GDI+) Could
713 * Allow Code Execution (833987)
714 * Enabled with +inspect-jpegs in actions files.
716 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
717 * - Fixed usage info for non-unix platforms
718 * - Fixed small cmdline parsing bug
720 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
721 * Diagnostics: Now preserve the returncode of pthread_create
722 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
724 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
725 * No longer ignore SIGCHLD. Fixes bug #769381
727 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
728 * Close jarfile on graceful exit, remove stray line
730 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
731 * Cosmetics: Killed a warning, a typo and an allocation left at exit
733 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
734 * No longer rely on non-POSIX.1 extensions of getcwd().
737 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
738 * Replaced setenv() by posix-compliant putenv()
739 * Thanks to Neil McCalden (nmcc AT users.sf.net).
741 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
742 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
744 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
745 * Clean-up and extension of improvements for forked mode:
746 * - Child's return code now consists of flags RC_FLAG_*
747 * - Reporting toggle to parent now properly #ifdef'ed
748 * - Children now report blocking to parent. This enables
749 * statistics in forked mode
751 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
752 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
753 * Children now report having been toggled through _exit(17), parents
754 * watch for that code and toggle themselves if found.
756 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
757 * Wrapping all *_r functions (the non-_r versions of them) with
758 * mutex semaphores for OSX. Hopefully this will take care of all
759 * of those pesky crash reports.
761 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
762 * Fixed two mostly harmless mem leaks
764 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
765 * Fix: Head of global clients list now initialized to NULL
767 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
768 * Added basic support for OPTIONS and TRACE HTTP methods:
769 * - New interceptor direct_response() added in chat().
770 * - sed() moved to earlier in the process, so that the
771 * Host: header is evaluated before actions and forwarding
774 * Revision 1.92 2002/05/08 16:00:46 oes
775 * Chat's buffer handling:
776 * - Fixed bug with unchecked out-of-mem conditions
777 * while reading client request & server headers
778 * - No longer predict if the buffer limit will be exceeded
779 * in the next read -- check add_to_iob's new
780 * return code. If buffer couldn't be extended
781 * (policy or out-of-mem) while
782 * - reading from client: abort
783 * - reading server headers: send error page
784 * - buffering server body for filter: flush,
785 * and if that fails: send error page
787 * Revision 1.91 2002/04/08 20:35:58 swa
790 * Revision 1.90 2002/04/02 14:57:28 oes
791 * Made sending wafers independent of FEATURE_COOKIE_JAR
793 * Revision 1.89 2002/03/31 17:18:59 jongfoster
794 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
796 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
797 * More compiler warning message maintenance
799 * Revision 1.87 2002/03/26 22:29:54 swa
800 * we have a new homepage!
802 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
803 * Workaround for closing the jarfile before load_config() comes around again
805 * Revision 1.85 2002/03/24 15:23:33 jongfoster
808 * Revision 1.84 2002/03/24 13:25:43 swa
809 * name change related issues
811 * Revision 1.83 2002/03/16 23:54:06 jongfoster
812 * Adding graceful termination feature, to help look for memory leaks.
813 * If you enable this (which, by design, has to be done by hand
814 * editing config.h) and then go to http://i.j.b/die, then the program
815 * will exit cleanly after the *next* request. It should free all the
816 * memory that was used.
818 * Revision 1.82 2002/03/13 00:27:05 jongfoster
821 * Revision 1.81 2002/03/12 01:42:50 oes
822 * Introduced modular filters
824 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
825 * OS/2 port maintenance:
826 * - Fixed EMX build - it had decayed a little
827 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
828 * substituted a memset for now.
830 * Revision 1.79 2002/03/09 20:03:52 jongfoster
831 * - Making various functions return int rather than size_t.
832 * (Undoing a recent change). Since size_t is unsigned on
833 * Windows, functions like read_socket that return -1 on
834 * error cannot return a size_t.
836 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
837 * crashes, and also frequently caused JB to jump to 100%
838 * CPU and stay there. (Because it thought it had just
839 * read ((unsigned)-1) == 4Gb of data...)
841 * - The signature of write_socket has changed, it now simply
842 * returns success=0/failure=nonzero.
844 * - Trying to get rid of a few warnings --with-debug on
845 * Windows, I've introduced a new type "jb_socket". This is
846 * used for the socket file descriptors. On Windows, this
847 * is SOCKET (a typedef for unsigned). Everywhere else, it's
848 * an int. The error value can't be -1 any more, so it's
849 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
850 * Windows it maps to the #define INVALID_SOCKET.)
852 * - The signature of bind_port has changed.
854 * Revision 1.78 2002/03/08 21:35:04 oes
855 * Added optional group supplement to --user option. Will now use default group of user if no group given
857 * Revision 1.77 2002/03/07 03:52:06 oes
858 * - Fixed compiler warnings etc
859 * - Improved handling of failed DNS lookups
861 * Revision 1.76 2002/03/06 22:54:35 jongfoster
862 * Automated function-comment nitpicking.
864 * Revision 1.75 2002/03/06 10:02:19 oes
865 * Fixed stupid bug when --user was not given
867 * Revision 1.74 2002/03/06 00:49:31 jongfoster
868 * Fixing warning on Windows
869 * Making #ifdefs that refer to the same variable consistently
870 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
872 * Revision 1.73 2002/03/05 23:57:30 hal9
873 * Stray character 's' on line 1618 was breaking build.
875 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
876 * - Re-enable OS/2 building after new parms were added
877 * - Fix false out of memory report when resolving CGI templates when no IP
878 * address is available of failed attempt (a la no such domain)
880 * Revision 1.71 2002/03/05 18:13:56 oes
881 * Added --user option
883 * Revision 1.70 2002/03/05 04:52:42 oes
884 * Deleted non-errlog debugging code
886 * Revision 1.69 2002/03/04 23:50:00 jongfoster
887 * Splitting off bind_port() call into bind_port_helper(), with
890 * Revision 1.68 2002/03/04 20:17:32 oes
893 * Revision 1.67 2002/03/04 18:18:57 oes
894 * - Removed _DEBUG mode
895 * - Cleand up cmdline parsing
896 * - Introduced --no-daemon, --pidfile options
897 * - Cleaned up signal handling:
898 * - Terminate cleanly on INT, TERM and ABRT
899 * - Schedule logfile for re-opening on HUP
900 * - Ignore CHLD and PIPE
901 * - Leave the rest with their default handlers
902 * - Uniform handler registration
903 * - Added usage() function
904 * - Played styleguide police
906 * Revision 1.66 2002/03/03 15:06:55 oes
907 * Re-enabled automatic config reloading
909 * Revision 1.65 2002/03/03 14:49:11 oes
910 * Fixed CLF logging: Now uses client's original HTTP request
912 * Revision 1.64 2002/03/03 09:18:03 joergs
913 * Made jumbjuster work on AmigaOS again.
915 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
916 * Clean up a little CRLF unpleasantness that suddenly appeared
918 * Revision 1.62 2002/02/20 23:17:23 jongfoster
919 * Detecting some out-of memory conditions and exiting with a log message.
921 * Revision 1.61 2002/01/17 21:01:52 jongfoster
922 * Moving all our URL and URL pattern parsing code to urlmatch.c.
924 * Revision 1.60 2001/12/30 14:07:32 steudten
925 * - Add signal handling (unix)
926 * - Add SIGHUP handler (unix)
927 * - Add creation of pidfile (unix)
928 * - Add action 'top' in rc file (RH)
929 * - Add entry 'SIGNALS' to manpage
930 * - Add exit message to logfile (unix)
932 * Revision 1.59 2001/12/13 14:07:18 oes
933 * Fixed Bug: 503 error page now sent OK
935 * Revision 1.58 2001/11/30 23:37:24 jongfoster
936 * Renaming the Win32 config file to config.txt - this is almost the
937 * same as the corresponding UNIX name "config"
939 * Revision 1.57 2001/11/16 00:47:43 jongfoster
940 * Changing the tty-disconnection code to use setsid().
942 * Revision 1.56 2001/11/13 20:20:54 jongfoster
943 * Tabs->spaces, fixing a bug with missing {} around an if()
945 * Revision 1.55 2001/11/13 20:14:53 jongfoster
946 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
948 * Revision 1.54 2001/11/07 00:03:14 steudten
949 * Give reliable return value if an error
950 * occurs not just 0 with new daemon mode.
952 * Revision 1.53 2001/11/05 21:41:43 steudten
953 * Add changes to be a real daemon just for unix os.
954 * (change cwd to /, detach from controlling tty, set
955 * process group and session leader to the own process.
957 * Add some fatal-error log message for failed malloc().
958 * Add '-d' if compiled with 'configure --with-debug' to
959 * enable debug output.
961 * Revision 1.52 2001/10/26 20:11:20 jongfoster
962 * Fixing type mismatch
964 * Revision 1.51 2001/10/26 17:38:28 oes
967 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
968 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
969 * threads to call select() simultaneously. So, it's time to do a real, live,
970 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
971 * (native). Both versions will work, but using __OS2__ offers multi-threading.
973 * Revision 1.49 2001/10/23 21:41:35 jongfoster
974 * Added call to initialize the (statically-allocated of course)
975 * "out of memory" CGI response.
977 * Revision 1.48 2001/10/10 19:56:46 jongfoster
978 * Moving some code that wasn't cookie-related out of an #ifdef
981 * Revision 1.47 2001/10/10 16:44:36 oes
982 * Added CONNECT destination port limitation check
984 * Revision 1.46 2001/10/08 15:17:41 oes
985 * Re-enabled SSL forwarding
987 * Revision 1.45 2001/10/07 15:42:11 oes
988 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
990 * Moved downgrading of the HTTP version from parse_http_request to
991 * chat(), since we can't decide if it is necessary before we have
992 * determined the actions for the URL. The HTTP command is now
993 * *always* re-built so the repairs need no longer be special-cased.
995 * filter_popups now gets a csp pointer so it can raise the new
996 * CSP_FLAG_MODIFIED flag.
1000 * Added configurable size limit for the IOB. If the IOB grows so
1001 * large that the next read would exceed the limit, the header
1002 * is generated, and the header & unfiltered buffer are flushed
1003 * to the client. Chat then continues in non-buffering,
1004 * non-filtering body mode.
1006 * Revision 1.44 2001/10/02 18:13:57 oes
1009 * Revision 1.43 2001/10/02 15:32:13 oes
1010 * Moved generation of hdr
1012 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
1013 * Cleaning up 2 compiler warnings on OS/2.
1015 * Revision 1.41 2001/09/16 17:05:14 jongfoster
1016 * Removing unused #include showarg.h
1018 * Revision 1.40 2001/09/16 15:41:45 jongfoster
1019 * Fixing signed/unsigned comparison warning.
1021 * Revision 1.39 2001/09/16 13:21:27 jongfoster
1022 * Changes to use new list functions.
1024 * Revision 1.38 2001/09/16 13:01:46 jongfoster
1025 * Removing redundant function call that zeroed zalloc()'d memory.
1027 * Revision 1.37 2001/09/10 11:12:24 oes
1028 * Deleted unused variable
1030 * Revision 1.36 2001/09/10 10:56:15 oes
1031 * Silenced compiler warnings
1033 * Revision 1.35 2001/07/31 14:44:22 oes
1034 * Deleted unused size parameter from filter_popups()
1036 * Revision 1.34 2001/07/30 22:08:36 jongfoster
1037 * Tidying up #defines:
1038 * - All feature #defines are now of the form FEATURE_xxx
1039 * - Permanently turned off WIN_GUI_EDIT
1040 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
1042 * Revision 1.33 2001/07/29 19:32:00 jongfoster
1043 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
1045 * Revision 1.32 2001/07/29 18:47:05 jongfoster
1046 * Adding missing #include "loadcfg.h"
1048 * Revision 1.31 2001/07/29 12:17:48 oes
1049 * Applied pthread fix by Paul Lieverse
1051 * Revision 1.30 2001/07/25 22:57:13 jongfoster
1052 * __BEOS__ no longer overrides FEATURE_PTHREAD.
1053 * This is because FEATURE_PTHREAD will soon be widely used, so I
1054 * want to keep it simple.
1056 * Revision 1.29 2001/07/24 12:47:06 oes
1057 * Applied BeOS support update by Eugenia
1059 * Revision 1.28 2001/07/23 13:26:12 oes
1060 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
1062 * Revision 1.27 2001/07/19 19:09:47 haroon
1063 * - Added code to take care of the situation where while processing the first
1064 * server response (which includes the server header), after finding the end
1065 * of the headers we were not looking past the end of the headers for
1066 * content modification. I enabled it for filter_popups.
1067 * Someone else should look to see if other similar operations should be
1068 * done to the discarded portion of the buffer.
1070 * Note 2001/07/20: No, the other content modification mechanisms will process
1071 * the whole iob later anyway. --oes
1073 * Revision 1.26 2001/07/18 12:31:36 oes
1076 * Revision 1.25 2001/07/15 19:43:49 jongfoster
1077 * Supports POSIX threads.
1078 * Also removed some unused #includes.
1080 * Revision 1.24 2001/07/13 14:00:40 oes
1081 * - Generic content modification scheme:
1082 * Each feature has its own applicability flag that is set
1083 * from csp->action->flags.
1084 * Replaced the "filtering" int flag , by a function pointer
1085 * "content_filter" to the function that will do the content
1086 * modification. If it is != NULL, the document will be buffered
1087 * and processed through *content_filter, which must set
1088 * csp->content_length and return a modified copy of the body
1089 * or return NULL (on failiure).
1090 * - Changed csp->is_text to the more generic bitmap csp->content_type
1091 * which can currently take the valued CT_TEXT or CT_GIF
1092 * - Reformatting etc
1093 * - Removed all #ifdef PCRS
1095 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
1096 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
1097 * the `block_acl' call. This prevents a compilation error when the user
1098 * does not wish to use the "ACL" feature.
1100 * Revision 1.22 2001/06/29 21:45:41 oes
1101 * Indentation, CRLF->LF, Tab-> Space
1103 * Revision 1.21 2001/06/29 13:29:36 oes
1104 * - Cleaned up, improved comments
1105 * - Unified all possible interceptors (CGI,
1106 * block, trust, fast_redirect) in one
1107 * place, with one (CGI) answer generation
1108 * mechansim. Much clearer now.
1109 * - Removed the GIF image generation, which
1110 * is now done in filters.c:block_url()
1111 * - Made error conditions like domain lookup
1112 * failiure or (various) problems while talking
1113 * to the server use cgi.c:error_response()
1114 * instead of generating HTML/HTTP in chat() (yuck!)
1115 * - Removed logentry from cancelled commit
1117 * Revision 1.20 2001/06/09 10:55:28 jongfoster
1118 * Changing BUFSIZ ==> BUFFER_SIZE
1120 * Revision 1.19 2001/06/07 23:12:52 jongfoster
1121 * Replacing function pointer in struct gateway with a directly
1122 * called function forwarded_connect().
1123 * Replacing struct gateway with struct forward_spec
1125 * Revision 1.18 2001/06/03 19:12:16 oes
1126 * introduced new cgi handling
1128 * Revision 1.17 2001/06/01 20:07:23 jongfoster
1129 * Now uses action +image-blocker{} rather than config->tinygif
1131 * Revision 1.16 2001/06/01 18:49:17 jongfoster
1132 * Replaced "list_share" with "list" - the tiny memory gain was not
1133 * worth the extra complexity.
1135 * Revision 1.15 2001/05/31 21:24:47 jongfoster
1136 * Changed "permission" to "action" throughout.
1137 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
1138 * Moved vanilla wafer check into chat(), since we must now
1139 * decide whether or not to add it based on the URL.
1141 * Revision 1.14 2001/05/29 20:14:01 joergs
1142 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
1145 * Revision 1.13 2001/05/29 09:50:24 jongfoster
1146 * Unified blocklist/imagelist/permissionslist.
1147 * File format is still under discussion, but the internal changes
1148 * are (mostly) done.
1150 * Also modified interceptor behaviour:
1151 * - We now intercept all URLs beginning with one of the following
1152 * prefixes (and *only* these prefixes):
1154 * * http://ijbswa.sf.net/config/
1155 * * http://ijbswa.sourceforge.net/config/
1156 * - New interceptors "home page" - go to http://i.j.b/ to see it.
1157 * - Internal changes so that intercepted and fast redirect pages
1158 * are not replaced with an image.
1159 * - Interceptors now have the option to send a binary page direct
1160 * to the client. (i.e. ijb-send-banner uses this)
1161 * - Implemented show-url-info interceptor. (Which is why I needed
1162 * the above interceptors changes - a typical URL is
1163 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
1164 * The previous mechanism would not have intercepted that, and
1165 * if it had been intercepted then it then it would have replaced
1166 * it with an image.)
1168 * Revision 1.12 2001/05/27 22:17:04 oes
1170 * - re_process_buffer no longer writes the modified buffer
1171 * to the client, which was very ugly. It now returns the
1172 * buffer, which it is then written by chat.
1174 * - content_length now adjusts the Content-Length: header
1175 * for modified documents rather than crunch()ing it.
1176 * (Length info in csp->content_length, which is 0 for
1177 * unmodified documents)
1179 * - For this to work, sed() is called twice when filtering.
1181 * Revision 1.11 2001/05/26 17:27:53 jongfoster
1182 * Added support for CLF and fixed LOG_LEVEL_LOG.
1183 * Also did CRLF->LF fix of my previous patch.
1185 * Revision 1.10 2001/05/26 15:26:15 jongfoster
1186 * ACL feature now provides more security by immediately dropping
1187 * connections from untrusted hosts.
1189 * Revision 1.9 2001/05/26 00:28:36 jongfoster
1190 * Automatic reloading of config file.
1191 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
1192 * Most of the global variables have been moved to a new
1193 * struct configuration_spec, accessed through csp->config->globalname
1194 * Most of the globals remaining are used by the Win32 GUI.
1196 * Revision 1.8 2001/05/25 22:43:18 jongfoster
1197 * Fixing minor memory leak and buffer overflow.
1199 * Revision 1.7 2001/05/25 22:34:30 jongfoster
1202 * Revision 1.6 2001/05/23 00:13:58 joergs
1203 * AmigaOS support fixed.
1205 * Revision 1.5 2001/05/22 18:46:04 oes
1207 * - Enabled filtering banners by size rather than URL
1208 * by adding patterns that replace all standard banner
1209 * sizes with the "Junkbuster" gif to the re_filterfile
1211 * - Enabled filtering WebBugs by providing a pattern
1212 * which kills all 1x1 images
1214 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
1215 * which is selected by the (nonstandard and therefore
1216 * capital) letter 'U' in the option string.
1217 * It causes the quantifiers to be ungreedy by default.
1218 * Appending a ? turns back to greedy (!).
1220 * - Added a new interceptor ijb-send-banner, which
1221 * sends back the "Junkbuster" gif. Without imagelist or
1222 * MSIE detection support, or if tinygif = 1, or the
1223 * URL isn't recognized as an imageurl, a lame HTML
1224 * explanation is sent instead.
1226 * - Added new feature, which permits blocking remote
1227 * script redirects and firing back a local redirect
1229 * The feature is conditionally compiled, i.e. it
1230 * can be disabled with --disable-fast-redirects,
1231 * plus it must be activated by a "fast-redirects"
1232 * line in the config file, has its own log level
1233 * and of course wants to be displayed by show-proxy-args
1234 * Note: Boy, all the #ifdefs in 1001 locations and
1235 * all the fumbling with configure.in and acconfig.h
1236 * were *way* more work than the feature itself :-(
1238 * - Because a generic redirect template was needed for
1239 * this, tinygif = 3 now uses the same.
1241 * - Moved GIFs, and other static HTTP response templates
1244 * - Some minor fixes
1246 * - Removed some >400 CRs again (Jon, you really worked
1249 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1250 * Made failure to bind() a fatal error.
1252 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1253 * Version 2.9.4 checkin.
1254 * - Merged popupfile and cookiefile, and added control over PCRS
1255 * filtering, in new "permissionsfile".
1256 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1257 * file error you now get a message box (in the Win32 GUI) rather
1258 * than the program exiting with no explanation.
1259 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1261 * - Removed tabs from "config"
1262 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1263 * - Bumped up version number.
1265 * Revision 1.2 2001/05/17 22:34:44 oes
1266 * - Added hint on GIF char array generation to jcc.c
1267 * - Cleaned CRLF's from the sources and related files
1268 * - Repaired logging for REF and FRC
1270 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1271 * Initial import of version 2.9.3 source tree
1274 *********************************************************************/
1280 #include <sys/types.h>
1289 # ifndef FEATURE_PTHREAD
1293 # include <windows.h>
1294 # include <process.h>
1295 # endif /* ndef FEATURE_PTHREAD */
1298 # ifndef _WIN_CONSOLE
1299 # include "w32log.h"
1300 # endif /* ndef _WIN_CONSOLE */
1301 # include "w32svrapi.h"
1303 #else /* ifndef _WIN32 */
1305 # if !defined (__OS2__)
1306 # include <unistd.h>
1307 # include <sys/wait.h>
1308 # endif /* ndef __OS2__ */
1309 # include <sys/time.h>
1310 # include <sys/stat.h>
1311 # include <sys/ioctl.h>
1314 #include <sys/termios.h>
1322 # include <signal.h>
1325 # include <socket.h> /* BeOS has select() for sockets only. */
1326 # include <OS.h> /* declarations for threads and stuff. */
1329 # if defined(__EMX__) || defined(__OS2__)
1330 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1335 #define bzero(B,N) memset(B,0x00,n)
1339 # include <select.h>
1344 #include "project.h"
1347 #include "filters.h"
1348 #include "loaders.h"
1349 #include "parsers.h"
1350 #include "miscutil.h"
1352 #include "jbsockets.h"
1353 #include "gateway.h"
1354 #include "actions.h"
1356 #include "loadcfg.h"
1357 #include "urlmatch.h"
1359 const char jcc_h_rcs[] = JCC_H_VERSION;
1360 const char project_h_rcs[] = PROJECT_H_VERSION;
1363 struct client_state clients[1];
1364 struct file_list files[1];
1366 #ifdef FEATURE_STATISTICS
1367 int urls_read = 0; /* total nr of urls read inc rejected */
1368 int urls_rejected = 0; /* total nr of urls rejected */
1369 #endif /* def FEATURE_STATISTICS */
1371 #ifdef FEATURE_GRACEFUL_TERMINATION
1372 int g_terminate = 0;
1375 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1376 static void sig_handler(int the_signal);
1378 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1379 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1380 static jb_err get_server_headers(struct client_state *csp);
1381 static const char *crunch_reason(const struct http_response *rsp);
1382 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1383 static char *get_request_line(struct client_state *csp);
1384 static jb_err receive_client_request(struct client_state *csp);
1385 static jb_err parse_client_request(struct client_state *csp);
1386 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1387 static jb_err change_request_destination(struct client_state *csp);
1388 static void chat(struct client_state *csp);
1389 static void serve(struct client_state *csp);
1390 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1391 static void usage(const char *myname);
1393 static void initialize_mutexes(void);
1394 static jb_socket bind_port_helper(struct configuration_spec *config);
1395 static void listen_loop(void);
1398 void serve(struct client_state *csp);
1399 #else /* ifndef AMIGA */
1400 static void serve(struct client_state *csp);
1401 #endif /* def AMIGA */
1404 static int32 server_thread(void *data);
1405 #endif /* def __BEOS__ */
1408 #define sleep(N) Sleep(((N) * 1000))
1412 #define sleep(N) DosSleep(((N) * 100))
1415 #ifdef MUTEX_LOCKS_AVAILABLE
1417 * XXX: Does the locking stuff really belong in this file?
1419 privoxy_mutex_t log_mutex;
1420 privoxy_mutex_t log_init_mutex;
1421 privoxy_mutex_t connection_reuse_mutex;
1423 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1424 privoxy_mutex_t resolver_mutex;
1425 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1427 #ifndef HAVE_GMTIME_R
1428 privoxy_mutex_t gmtime_mutex;
1429 #endif /* ndef HAVE_GMTIME_R */
1431 #ifndef HAVE_LOCALTIME_R
1432 privoxy_mutex_t localtime_mutex;
1433 #endif /* ndef HAVE_GMTIME_R */
1436 privoxy_mutex_t rand_mutex;
1437 #endif /* ndef HAVE_RANDOM */
1439 #endif /* def MUTEX_LOCKS_AVAILABLE */
1442 const char *basedir = NULL;
1443 const char *pidfile = NULL;
1444 static int received_hup_signal = 0;
1445 #endif /* defined unix */
1447 /* HTTP snipplets. */
1448 static const char CSUCCEED[] =
1449 "HTTP/1.0 200 Connection established\r\n"
1450 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1452 static const char CHEADER[] =
1453 "HTTP/1.0 400 Invalid header received from client\r\n"
1454 "Proxy-Agent: Privoxy " VERSION "\r\n"
1455 "Content-Type: text/plain\r\n"
1456 "Connection: close\r\n\r\n"
1457 "Invalid header received from client.\r\n";
1459 static const char FTP_RESPONSE[] =
1460 "HTTP/1.0 400 Invalid request received from client\r\n"
1461 "Content-Type: text/plain\r\n"
1462 "Connection: close\r\n\r\n"
1463 "Invalid request. Privoxy doesn't support FTP.\r\n";
1465 static const char GOPHER_RESPONSE[] =
1466 "HTTP/1.0 400 Invalid request received from client\r\n"
1467 "Content-Type: text/plain\r\n"
1468 "Connection: close\r\n\r\n"
1469 "Invalid request. Privoxy doesn't support gopher.\r\n";
1471 /* XXX: should be a template */
1472 static const char MISSING_DESTINATION_RESPONSE[] =
1473 "HTTP/1.0 400 Bad request received from client\r\n"
1474 "Proxy-Agent: Privoxy " VERSION "\r\n"
1475 "Content-Type: text/plain\r\n"
1476 "Connection: close\r\n\r\n"
1477 "Bad request. Privoxy was unable to extract the destination.\r\n";
1479 /* XXX: should be a template */
1480 static const char NO_SERVER_DATA_RESPONSE[] =
1481 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1482 "Proxy-Agent: Privoxy " VERSION "\r\n"
1483 "Content-Type: text/plain\r\n"
1484 "Connection: close\r\n\r\n"
1485 "Empty server or forwarder response.\r\n"
1486 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1488 /* XXX: should be a template */
1489 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1490 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1491 "Proxy-Agent: Privoxy " VERSION "\r\n"
1492 "Content-Type: text/plain\r\n"
1493 "Connection: close\r\n\r\n"
1494 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1497 /* XXX: should be a template */
1498 static const char NULL_BYTE_RESPONSE[] =
1499 "HTTP/1.0 400 Bad request received from client\r\n"
1500 "Proxy-Agent: Privoxy " VERSION "\r\n"
1501 "Content-Type: text/plain\r\n"
1502 "Connection: close\r\n\r\n"
1503 "Bad request. Null byte(s) before end of request.\r\n";
1506 /* XXX: should be a template */
1507 static const char MESSED_UP_REQUEST_RESPONSE[] =
1508 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1509 "Proxy-Agent: Privoxy " VERSION "\r\n"
1510 "Content-Type: text/plain\r\n"
1511 "Connection: close\r\n\r\n"
1512 "Bad request. Messed up with header filters.\r\n";
1514 static const char TOO_MANY_CONNECTIONS_RESPONSE[] =
1515 "HTTP/1.0 503 Too many open connections\r\n"
1516 "Proxy-Agent: Privoxy " VERSION "\r\n"
1517 "Content-Type: text/plain\r\n"
1518 "Connection: close\r\n\r\n"
1519 "Maximum number of open connections reached.\r\n";
1521 /* XXX: should be a template */
1522 static const char CONNECTION_TIMEOUT_RESPONSE[] =
1523 "HTTP/1.0 504 Connection timeout\r\n"
1524 "Proxy-Agent: Privoxy " VERSION "\r\n"
1525 "Content-Type: text/plain\r\n"
1526 "Connection: close\r\n\r\n"
1527 "The connection timed out.\r\n";
1529 /* A function to crunch a response */
1530 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1532 /* Crunch function flags */
1533 #define CF_NO_FLAGS 0
1534 /* Cruncher applies to forced requests as well */
1535 #define CF_IGNORE_FORCE 1
1536 /* Crunched requests are counted for the block statistics */
1537 #define CF_COUNT_AS_REJECT 2
1539 /* A crunch function and its flags */
1542 const crunch_func_ptr cruncher;
1546 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1548 /* Complete list of cruncher functions */
1549 static const struct cruncher crunchers_all[] = {
1550 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1551 { block_url, CF_COUNT_AS_REJECT },
1552 #ifdef FEATURE_TRUST
1553 { trust_url, CF_COUNT_AS_REJECT },
1554 #endif /* def FEATURE_TRUST */
1555 { redirect_url, CF_NO_FLAGS },
1556 { dispatch_cgi, CF_IGNORE_FORCE},
1560 /* Light version, used after tags are applied */
1561 static const struct cruncher crunchers_light[] = {
1562 { block_url, CF_COUNT_AS_REJECT },
1563 { redirect_url, CF_NO_FLAGS },
1569 * XXX: Don't we really mean
1575 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1576 /*********************************************************************
1578 * Function : sig_handler
1580 * Description : Signal handler for different signals.
1581 * Exit gracefully on TERM and INT
1582 * or set a flag that will cause the errlog
1583 * to be reopened by the main thread on HUP.
1586 * 1 : the_signal = the signal cause this function to call
1590 *********************************************************************/
1591 static void sig_handler(int the_signal)
1597 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1609 received_hup_signal = 1;
1615 * We shouldn't be here, unless we catch signals
1616 * in main() that we can't handle here!
1618 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1626 /*********************************************************************
1628 * Function : client_protocol_is_unsupported
1630 * Description : Checks if the client used a known unsupported
1631 * protocol and deals with it by sending an error
1635 * 1 : csp = Current client state (buffers, headers, etc...)
1636 * 2 : req = the first request line send by the client
1638 * Returns : TRUE if an error response has been generated, or
1639 * FALSE if the request doesn't look invalid.
1641 *********************************************************************/
1642 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1645 * If it's a FTP or gopher request, we don't support it.
1647 * These checks are better than nothing, but they might
1648 * not work in all configurations and some clients might
1649 * have problems digesting the answer.
1651 * They should, however, never cause more problems than
1652 * Privoxy's old behaviour (returning the misleading HTML
1655 * "Could not resolve http://(ftp|gopher)://example.org").
1657 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1659 const char *response = NULL;
1660 const char *protocol = NULL;
1662 if (!strncmpic(req, "GET ftp://", 10))
1664 response = FTP_RESPONSE;
1669 response = GOPHER_RESPONSE;
1670 protocol = "GOPHER";
1672 log_error(LOG_LEVEL_ERROR,
1673 "%s tried to use Privoxy as %s proxy: %s",
1674 csp->ip_addr_str, protocol, req);
1675 log_error(LOG_LEVEL_CLF,
1676 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1678 write_socket(csp->cfd, response, strlen(response));
1687 /*********************************************************************
1689 * Function : get_request_destination_elsewhere
1691 * Description : If the client's request was redirected into
1692 * Privoxy without the client's knowledge,
1693 * the request line lacks the destination host.
1695 * This function tries to get it elsewhere,
1696 * provided accept-intercepted-requests is enabled.
1698 * "Elsewhere" currently only means "Host: header",
1699 * but in the future we may ask the redirecting
1700 * packet filter to look the destination up.
1702 * If the destination stays unknown, an error
1703 * response is send to the client and headers
1704 * are freed so that chat() can return directly.
1707 * 1 : csp = Current client state (buffers, headers, etc...)
1708 * 2 : headers = a header list
1710 * Returns : JB_ERR_OK if the destination is now known, or
1711 * JB_ERR_PARSE if it isn't.
1713 *********************************************************************/
1714 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1718 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1720 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1721 " Privoxy isn't configured to accept intercepted requests.",
1722 csp->ip_addr_str, csp->http->cmd);
1723 /* XXX: Use correct size */
1724 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1725 csp->ip_addr_str, csp->http->cmd);
1727 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1728 destroy_list(headers);
1730 return JB_ERR_PARSE;
1732 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1734 /* Split the domain we just got for pattern matching */
1735 init_domain_components(csp->http);
1741 /* We can't work without destination. Go spread the news.*/
1743 req = list_to_text(headers);
1745 /* XXX: Use correct size */
1746 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1747 csp->ip_addr_str, csp->http->cmd);
1748 log_error(LOG_LEVEL_ERROR,
1749 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1750 csp->ip_addr_str, csp->http->cmd, req);
1753 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1754 destroy_list(headers);
1756 return JB_ERR_PARSE;
1759 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1760 * to get the destination IP address, use it as host directly
1761 * or do a reverse DNS lookup first.
1766 /*********************************************************************
1768 * Function : get_server_headers
1770 * Description : Parses server headers in iob and fills them
1771 * into csp->headers so that they can later be
1775 * 1 : csp = Current client state (buffers, headers, etc...)
1777 * Returns : JB_ERR_OK if everything went fine, or
1778 * JB_ERR_PARSE if the headers were incomplete.
1780 *********************************************************************/
1781 static jb_err get_server_headers(struct client_state *csp)
1783 int continue_hack_in_da_house = 0;
1786 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1791 * continue hack in da house. Ignore the ending of
1792 * this head and continue enlisting header lines.
1793 * The reason is described below.
1795 enlist(csp->headers, "");
1796 continue_hack_in_da_house = 0;
1799 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1802 * It's a bodyless continue response, don't
1803 * stop header parsing after reaching its end.
1805 * As a result Privoxy will concatenate the
1806 * next response's head and parse and deliver
1807 * the headers as if they belonged to one request.
1809 * The client will separate them because of the
1810 * empty line between them.
1812 * XXX: What we're doing here is clearly against
1813 * the intended purpose of the continue header,
1814 * and under some conditions (HTTP/1.0 client request)
1815 * it's a standard violation.
1817 * Anyway, "sort of against the spec" is preferable
1818 * to "always getting confused by Continue responses"
1819 * (Privoxy's behaviour before this hack was added)
1821 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1822 continue_hack_in_da_house = 1;
1824 else if (*header == '\0')
1827 * If the header is empty, but the Continue hack
1828 * isn't active, we can assume that we reached the
1829 * end of the buffer before we hit the end of the
1832 * Inform the caller an let it decide how to handle it.
1834 return JB_ERR_PARSE;
1837 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1840 * XXX: Should we quit the request and return a
1841 * out of memory error page instead?
1843 log_error(LOG_LEVEL_ERROR,
1844 "Out of memory while enlisting server headers. %s lost.",
1854 /*********************************************************************
1856 * Function : crunch_reason
1858 * Description : Translates the crunch reason code into a string.
1861 * 1 : rsp = a http_response
1863 * Returns : A string with the crunch reason or an error description.
1865 *********************************************************************/
1866 static const char *crunch_reason(const struct http_response *rsp)
1868 char * reason = NULL;
1870 assert(rsp != NULL);
1873 return "Internal error while searching for crunch reason";
1876 switch (rsp->reason)
1878 case RSP_REASON_UNSUPPORTED:
1879 reason = "Unsupported HTTP feature";
1881 case RSP_REASON_BLOCKED:
1884 case RSP_REASON_UNTRUSTED:
1885 reason = "Untrusted";
1887 case RSP_REASON_REDIRECTED:
1888 reason = "Redirected";
1890 case RSP_REASON_CGI_CALL:
1891 reason = "CGI Call";
1893 case RSP_REASON_NO_SUCH_DOMAIN:
1894 reason = "DNS failure";
1896 case RSP_REASON_FORWARDING_FAILED:
1897 reason = "Forwarding failed";
1899 case RSP_REASON_CONNECT_FAILED:
1900 reason = "Connection failure";
1902 case RSP_REASON_OUT_OF_MEMORY:
1903 reason = "Out of memory (may mask other reasons)";
1906 reason = "No reason recorded";
1914 /*********************************************************************
1916 * Function : send_crunch_response
1918 * Description : Delivers already prepared response for
1919 * intercepted requests, logs the interception
1920 * and frees the response.
1923 * 1 : csp = Current client state (buffers, headers, etc...)
1924 * 1 : rsp = Fully prepared response. Will be freed on exit.
1926 * Returns : Nothing.
1928 *********************************************************************/
1929 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1931 const struct http_request *http = csp->http;
1932 char status_code[4];
1934 assert(rsp != NULL);
1935 assert(rsp->head != NULL);
1940 * Not supposed to happen. If it does
1941 * anyway, treat it as an unknown error.
1943 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1944 /* return code doesn't matter */
1949 /* If rsp is still NULL, we have serious internal problems. */
1950 log_error(LOG_LEVEL_FATAL,
1951 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1955 * Extract the status code from the actual head
1956 * that was send to the client. It is the only
1957 * way to get it right for all requests, including
1958 * the fixed ones for out-of-memory problems.
1960 * A head starts like this: 'HTTP/1.1 200...'
1964 status_code[0] = rsp->head[9];
1965 status_code[1] = rsp->head[10];
1966 status_code[2] = rsp->head[11];
1967 status_code[3] = '\0';
1969 /* Write the answer to the client */
1970 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1971 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1973 /* There is nothing we can do about it. */
1974 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1977 /* Log that the request was crunched and why. */
1978 log_error(LOG_LEVEL_CRUNCH, "%s: %s", crunch_reason(rsp), http->url);
1979 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %u",
1980 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1982 /* Clean up and return */
1983 if (cgi_error_memory() != rsp)
1985 free_http_response(rsp);
1992 /*********************************************************************
1994 * Function : request_contains_null_bytes
1996 * Description : Checks for NULL bytes in the request and sends
1997 * an error message to the client if any were found.
1999 * XXX: currently not used, see comment in chat().
2002 * 1 : csp = Current client state (buffers, headers, etc...)
2003 * 2 : buf = Data from the client's request to check.
2004 * 3 : len = The data length.
2006 * Returns : TRUE if the request contained one or more NULL bytes, or
2009 *********************************************************************/
2010 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
2012 size_t c_len; /* Request lenght when treated as C string */
2014 c_len = strlen(buf);
2019 * Null byte(s) found. Log the request,
2020 * return an error response and hang up.
2022 size_t tmp_len = c_len;
2027 * Replace NULL byte(s) with '°' characters
2028 * so the request can be logged as string.
2029 * XXX: Is there a better replacement character?
2032 tmp_len += strlen(buf+tmp_len);
2033 } while (tmp_len < len);
2035 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
2036 "(length=%d, strlen=%u).", csp->ip_addr_str, len, c_len);
2037 log_error(LOG_LEVEL_HEADER,
2038 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
2040 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
2042 /* XXX: Log correct size */
2043 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2053 /*********************************************************************
2055 * Function : crunch_response_triggered
2057 * Description : Checks if the request has to be crunched,
2058 * and delivers the crunch response if necessary.
2061 * 1 : csp = Current client state (buffers, headers, etc...)
2062 * 2 : crunchers = list of cruncher functions to run
2064 * Returns : TRUE if the request was answered with a crunch response
2067 *********************************************************************/
2068 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
2070 struct http_response *rsp = NULL;
2071 const struct cruncher *c;
2074 * If CGI request crunching is disabled,
2075 * check the CGI dispatcher out of order to
2076 * prevent unintentional blocks or redirects.
2078 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
2079 && (NULL != (rsp = dispatch_cgi(csp))))
2081 /* Deliver, log and free the interception response. */
2082 send_crunch_response(csp, rsp);
2086 for (c = crunchers; c->cruncher != NULL; c++)
2089 * Check the cruncher if either Privoxy is toggled
2090 * on and the request isn't forced, or if the cruncher
2091 * applies to forced requests as well.
2093 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
2094 !(csp->flags & CSP_FLAG_FORCED)) ||
2095 (c->flags & CF_IGNORE_FORCE))
2097 rsp = c->cruncher(csp);
2100 /* Deliver, log and free the interception response. */
2101 send_crunch_response(csp, rsp);
2102 #ifdef FEATURE_STATISTICS
2103 if (c->flags & CF_COUNT_AS_REJECT)
2105 csp->flags |= CSP_FLAG_REJECTED;
2107 #endif /* def FEATURE_STATISTICS */
2118 /*********************************************************************
2120 * Function : build_request_line
2122 * Description : Builds the HTTP request line.
2124 * If a HTTP forwarder is used it expects the whole URL,
2125 * web servers only get the path.
2128 * 1 : csp = Current client state (buffers, headers, etc...)
2129 * 2 : fwd = The forwarding spec used for the request
2130 * XXX: Should use http->fwd instead.
2131 * 3 : request_line = The old request line which will be replaced.
2133 * Returns : Nothing. Terminates in case of memory problems.
2135 *********************************************************************/
2136 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
2138 struct http_request *http = csp->http;
2140 assert(http->ssl == 0);
2143 * Downgrade http version from 1.1 to 1.0
2144 * if +downgrade action applies.
2146 if ( (csp->action->flags & ACTION_DOWNGRADE)
2147 && (!strcmpic(http->ver, "HTTP/1.1")))
2150 http->ver = strdup("HTTP/1.0");
2152 if (http->ver == NULL)
2154 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
2159 * Rebuild the request line.
2161 freez(*request_line);
2162 *request_line = strdup(http->gpc);
2163 string_append(request_line, " ");
2165 if (fwd->forward_host)
2167 string_append(request_line, http->url);
2171 string_append(request_line, http->path);
2173 string_append(request_line, " ");
2174 string_append(request_line, http->ver);
2176 if (*request_line == NULL)
2178 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
2180 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
2184 /*********************************************************************
2186 * Function : change_request_destination
2188 * Description : Parse a (rewritten) request line and regenerate
2189 * the http request data.
2192 * 1 : csp = Current client state (buffers, headers, etc...)
2194 * Returns : Forwards the parse_http_request() return code.
2195 * Terminates in case of memory problems.
2197 *********************************************************************/
2198 static jb_err change_request_destination(struct client_state *csp)
2200 struct http_request *http = csp->http;
2203 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
2204 free_http_request(http);
2205 err = parse_http_request(csp->headers->first->str, http);
2206 if (JB_ERR_OK != err)
2208 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
2209 jb_err_to_string(err));
2213 /* XXX: ocmd is a misleading name */
2214 http->ocmd = strdup(http->cmd);
2215 if (http->ocmd == NULL)
2217 log_error(LOG_LEVEL_FATAL,
2218 "Out of memory copying rewritten HTTP request line");
2226 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2227 /*********************************************************************
2229 * Function : server_response_is_complete
2231 * Description : Determines whether we should stop reading
2232 * from the server socket.
2235 * 1 : csp = Current client state (buffers, headers, etc...)
2236 * 2 : content_length = Length of content received so far.
2238 * Returns : TRUE if the response is complete,
2241 *********************************************************************/
2242 static int server_response_is_complete(struct client_state *csp,
2243 unsigned long long content_length)
2245 int content_length_known = !!(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2247 if (!strcmpic(csp->http->gpc, "HEAD"))
2250 * "HEAD" implies no body, we are thus expecting
2251 * no content. XXX: incomplete "list" of methods?
2253 csp->expected_content_length = 0;
2254 content_length_known = TRUE;
2257 if (csp->http->status == 304)
2260 * Expect no body. XXX: incomplete "list" of status codes?
2262 csp->expected_content_length = 0;
2263 content_length_known = TRUE;
2266 return (content_length_known && ((0 == csp->expected_content_length)
2267 || (csp->expected_content_length <= content_length)));
2271 /*********************************************************************
2273 * Function : wait_for_alive_connections
2275 * Description : Waits for alive connections to timeout.
2281 *********************************************************************/
2282 static void wait_for_alive_connections()
2284 int connections_alive = close_unusable_connections();
2286 while (0 < connections_alive)
2288 log_error(LOG_LEVEL_CONNECT,
2289 "Waiting for %d connections to timeout.",
2292 connections_alive = close_unusable_connections();
2295 log_error(LOG_LEVEL_CONNECT, "No connections to wait for left.");
2300 /*********************************************************************
2302 * Function : save_connection_destination
2304 * Description : Remembers a connection for reuse later on.
2307 * 1 : sfd = Open socket to remember.
2308 * 2 : http = The destination for the connection.
2309 * 3 : fwd = The forwarder settings used.
2310 * 3 : server_connection = storage.
2314 *********************************************************************/
2315 void save_connection_destination(jb_socket sfd,
2316 const struct http_request *http,
2317 const struct forward_spec *fwd,
2318 struct reusable_connection *server_connection)
2320 assert(sfd != JB_INVALID_SOCKET);
2321 assert(NULL != http->host);
2322 server_connection->host = strdup(http->host);
2323 if (NULL == server_connection->host)
2325 log_error(LOG_LEVEL_FATAL, "Out of memory saving socket.");
2327 server_connection->port = http->port;
2329 assert(NULL != fwd);
2330 assert(server_connection->gateway_host == NULL);
2331 assert(server_connection->gateway_port == 0);
2332 assert(server_connection->forwarder_type == 0);
2333 assert(server_connection->forward_host == NULL);
2334 assert(server_connection->forward_port == 0);
2336 server_connection->forwarder_type = fwd->type;
2337 if (NULL != fwd->gateway_host)
2339 server_connection->gateway_host = strdup(fwd->gateway_host);
2340 if (NULL == server_connection->gateway_host)
2342 log_error(LOG_LEVEL_FATAL, "Out of memory saving gateway_host.");
2347 server_connection->gateway_host = NULL;
2349 server_connection->gateway_port = fwd->gateway_port;
2351 if (NULL != fwd->forward_host)
2353 server_connection->forward_host = strdup(fwd->forward_host);
2354 if (NULL == server_connection->forward_host)
2356 log_error(LOG_LEVEL_FATAL, "Out of memory saving forward_host.");
2361 server_connection->forward_host = NULL;
2363 server_connection->forward_port = fwd->forward_port;
2365 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2368 /*********************************************************************
2370 * Function : mark_server_socket_tainted
2372 * Description : Makes sure we don't reuse a server socket
2373 * (if we didn't read everything the server sent
2374 * us reusing the socket would lead to garbage).
2377 * 1 : csp = Current client state (buffers, headers, etc...)
2381 *********************************************************************/
2382 static void mark_server_socket_tainted(struct client_state *csp)
2384 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2386 log_error(LOG_LEVEL_CONNECT, "Unsetting keep-alive flag.");
2387 csp->flags &= ~CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE;
2391 /*********************************************************************
2393 * Function : get_request_line
2395 * Description : Read the client request line.
2398 * 1 : csp = Current client state (buffers, headers, etc...)
2400 * Returns : Pointer to request line or NULL in case of errors.
2402 *********************************************************************/
2403 static char *get_request_line(struct client_state *csp)
2405 char buf[BUFFER_SIZE];
2406 char *request_line = NULL;
2409 memset(buf, 0, sizeof(buf));
2413 if (!data_is_available(csp->cfd, csp->config->socket_timeout))
2415 log_error(LOG_LEVEL_ERROR,
2416 "Stopped waiting for the request line.");
2417 write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE,
2418 strlen(CONNECTION_TIMEOUT_RESPONSE));
2422 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2424 if (len <= 0) return NULL;
2427 * If there is no memory left for buffering the
2428 * request, there is nothing we can do but hang up
2430 if (add_to_iob(csp, buf, len))
2435 request_line = get_header(csp->iob);
2437 } while ((NULL != request_line) && ('\0' == *request_line));
2439 return request_line;
2444 /*********************************************************************
2446 * Function : receive_client_request
2448 * Description : Read the client's request (more precisely the
2449 * client headers) and answer it if necessary.
2451 * Note that since we're not using select() we could get
2452 * blocked here if a client connected, then didn't say
2456 * 1 : csp = Current client state (buffers, headers, etc...)
2458 * Returns : JB_ERR_OK, JB_ERR_PARSE or JB_ERR_MEMORY
2460 *********************************************************************/
2461 static jb_err receive_client_request(struct client_state *csp)
2463 char buf[BUFFER_SIZE];
2466 struct http_request *http;
2470 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2471 struct list header_list;
2472 struct list *headers = &header_list;
2476 memset(buf, 0, sizeof(buf));
2478 req = get_request_line(csp);
2481 return JB_ERR_PARSE;
2483 assert(*req != '\0');
2485 if (client_protocol_is_unsupported(csp, req))
2487 return JB_ERR_PARSE;
2490 #ifdef FEATURE_FORCE_LOAD
2492 * If this request contains the FORCE_PREFIX and blocks
2493 * aren't enforced, get rid of it and set the force flag.
2495 if (strstr(req, FORCE_PREFIX))
2497 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2499 log_error(LOG_LEVEL_FORCE,
2500 "Ignored force prefix in request: \"%s\".", req);
2504 strclean(req, FORCE_PREFIX);
2505 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2506 csp->flags |= CSP_FLAG_FORCED;
2509 #endif /* def FEATURE_FORCE_LOAD */
2511 err = parse_http_request(req, http);
2513 if (JB_ERR_OK != err)
2515 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2516 /* XXX: Use correct size */
2517 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2518 log_error(LOG_LEVEL_ERROR,
2519 "Couldn't parse request line received from %s: %s",
2520 csp->ip_addr_str, jb_err_to_string(err));
2522 free_http_request(http);
2523 return JB_ERR_PARSE;
2526 /* grab the rest of the client's headers */
2530 p = get_header(csp->iob);
2534 /* There are no additional headers to read. */
2541 * We didn't receive a complete header
2542 * line yet, get the rest of it.
2544 if (!data_is_available(csp->cfd, csp->config->socket_timeout))
2546 log_error(LOG_LEVEL_ERROR,
2547 "Stopped grabbing the client headers.");
2548 return JB_ERR_PARSE;
2551 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2554 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2555 destroy_list(headers);
2556 return JB_ERR_PARSE;
2559 if (add_to_iob(csp, buf, len))
2562 * If there is no memory left for buffering the
2563 * request, there is nothing we can do but hang up
2565 destroy_list(headers);
2566 return JB_ERR_MEMORY;
2572 * We were able to read a complete
2573 * header and can finaly enlist it.
2580 if (http->host == NULL)
2583 * If we still don't know the request destination,
2584 * the request is invalid or the client uses
2585 * Privoxy without its knowledge.
2587 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2590 * Our attempts to get the request destination
2591 * elsewhere failed or Privoxy is configured
2592 * to only accept proxy requests.
2594 * An error response has already been send
2595 * and we're done here.
2597 return JB_ERR_PARSE;
2602 * Determine the actions for this URL
2604 #ifdef FEATURE_TOGGLE
2605 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2607 /* Most compatible set of actions (i.e. none) */
2608 init_current_action(csp->action);
2611 #endif /* ndef FEATURE_TOGGLE */
2613 get_url_actions(csp, http);
2617 * Save a copy of the original request for logging
2619 http->ocmd = strdup(http->cmd);
2620 if (http->ocmd == NULL)
2622 log_error(LOG_LEVEL_FATAL,
2623 "Out of memory copying HTTP request line");
2625 enlist(csp->headers, http->cmd);
2627 /* Append the previously read headers */
2628 list_append_list_unique(csp->headers, headers);
2629 destroy_list(headers);
2636 /*********************************************************************
2638 * Function : parse_client_request
2640 * Description : Parses the client's request and decides what to do
2643 * Note that since we're not using select() we could get
2644 * blocked here if a client connected, then didn't say
2648 * 1 : csp = Current client state (buffers, headers, etc...)
2650 * Returns : JB_ERR_OK or JB_ERR_PARSE
2652 *********************************************************************/
2653 static jb_err parse_client_request(struct client_state *csp)
2655 struct http_request *http = csp->http;
2658 err = sed(csp, FILTER_CLIENT_HEADERS);
2659 if (JB_ERR_OK != err)
2661 /* XXX: Should be handled in sed(). */
2662 assert(err == JB_ERR_PARSE);
2663 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers.");
2665 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2667 /* Check request line for rewrites. */
2668 if ((NULL == csp->headers->first->str)
2669 || (strcmp(http->cmd, csp->headers->first->str) &&
2670 (JB_ERR_OK != change_request_destination(csp))))
2673 * A header filter broke the request line - bail out.
2675 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2676 /* XXX: Use correct size */
2677 log_error(LOG_LEVEL_CLF,
2678 "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2679 log_error(LOG_LEVEL_ERROR,
2680 "Invalid request line after applying header filters.");
2681 free_http_request(http);
2683 return JB_ERR_PARSE;
2691 /*********************************************************************
2695 * Description : Once a connection to the client has been accepted,
2696 * this function is called (via serve()) to handle the
2697 * main business of the communication. When this
2698 * function returns, the caller must close the client
2701 * FIXME: chat is nearly thousand lines long.
2705 * 1 : csp = Current client state (buffers, headers, etc...)
2707 * Returns : Nothing.
2709 *********************************************************************/
2710 static void chat(struct client_state *csp)
2712 char buf[BUFFER_SIZE];
2719 int ms_iis5_hack = 0;
2720 unsigned long long byte_count = 0;
2721 int forwarded_connect_retries = 0;
2722 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2723 const struct forward_spec *fwd;
2724 struct http_request *http;
2725 long len = 0; /* for buffer sizes (and negative error codes) */
2727 /* Function that does the content filtering for the current request */
2728 filter_function_ptr content_filter = NULL;
2730 /* Skeleton for HTTP response, if we should intercept the request */
2731 struct http_response *rsp;
2732 struct timeval timeout;
2734 memset(buf, 0, sizeof(buf));
2738 if (receive_client_request(csp) != JB_ERR_OK)
2742 if (parse_client_request(csp) != JB_ERR_OK)
2747 /* decide how to route the HTTP request */
2748 fwd = forward_url(csp, http);
2751 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2752 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2757 * build the http request to send to the server
2758 * we have to do one of the following:
2760 * create = use the original HTTP request to create a new
2761 * HTTP request that has either the path component
2762 * without the http://domainspec (w/path) or the
2763 * full orininal URL (w/url)
2764 * Note that the path and/or the HTTP version may
2765 * have been altered by now.
2767 * connect = Open a socket to the host:port of the server
2768 * and short-circuit server and client socket.
2770 * pass = Pass the request unchanged if forwarding a CONNECT
2771 * request to a parent proxy. Note that we'll be sending
2772 * the CFAIL message ourselves if connecting to the parent
2773 * fails, but we won't send a CSUCCEED message if it works,
2774 * since that would result in a double message (ours and the
2775 * parent's). After sending the request to the parent, we simply
2778 * here's the matrix:
2781 * +--------+--------+
2783 * 0 | create | connect|
2785 * Forwarding +--------+--------+
2787 * 1 | create | pass |
2789 * +--------+--------+
2793 if (http->ssl && connect_port_is_forbidden(csp))
2795 const char *acceptable_connect_ports =
2796 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2797 assert(NULL != acceptable_connect_ports);
2798 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2799 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2800 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2801 csp->action->flags |= ACTION_BLOCK;
2807 freez(csp->headers->first->str);
2808 build_request_line(csp, fwd, &csp->headers->first->str);
2812 * We have a request. Check if one of the crunchers wants it.
2814 if (crunch_response_triggered(csp, crunchers_all))
2817 * Yes. The client got the crunch response
2818 * and we are done here after cleaning up.
2820 /* XXX: why list_remove_all()? */
2821 list_remove_all(csp->headers);
2826 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2828 if (fwd->forward_host)
2830 log_error(LOG_LEVEL_CONNECT, "via [%s]:%d to: %s",
2831 fwd->forward_host, fwd->forward_port, http->hostport);
2835 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2838 /* here we connect to the server, gateway, or the forwarder */
2840 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2841 if ((csp->sfd != JB_INVALID_SOCKET)
2842 && socket_is_still_usable(csp->sfd)
2843 && connection_destination_matches(&csp->server_connection, http, fwd))
2845 log_error(LOG_LEVEL_CONNECT,
2846 "Reusing server socket %u. Opened for %s.",
2847 csp->sfd, csp->server_connection.host);
2851 if (csp->sfd != JB_INVALID_SOCKET)
2853 log_error(LOG_LEVEL_CONNECT,
2854 "Closing server socket %u. Opened for %s.",
2855 csp->sfd, csp->server_connection.host);
2856 close_socket(csp->sfd);
2857 mark_connection_closed(&csp->server_connection);
2859 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
2861 while ((csp->sfd = forwarded_connect(fwd, http, csp))
2862 && (errno == EINVAL)
2863 && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2865 log_error(LOG_LEVEL_ERROR,
2866 "failed request #%u to connect to %s. Trying again.",
2867 forwarded_connect_retries, http->hostport);
2870 if (csp->sfd == JB_INVALID_SOCKET)
2872 if (fwd->type != SOCKS_NONE)
2875 rsp = error_response(csp, "forwarding-failed", errno);
2877 else if (errno == EINVAL)
2879 rsp = error_response(csp, "no-such-domain", errno);
2883 rsp = error_response(csp, "connect-failed", errno);
2884 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2888 /* Write the answer to the client */
2891 send_crunch_response(csp, rsp);
2896 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2897 save_connection_destination(csp->sfd, http, fwd, &csp->server_connection);
2899 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
2901 hdr = list_to_text(csp->headers);
2904 /* FIXME Should handle error properly */
2905 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2907 list_remove_all(csp->headers);
2909 if (fwd->forward_host || (http->ssl == 0))
2912 * Write the client's (modified) header to the server
2913 * (along with anything else that may be in the buffer)
2915 if (write_socket(csp->sfd, hdr, strlen(hdr))
2916 || (flush_socket(csp->sfd, csp->iob) < 0))
2918 log_error(LOG_LEVEL_CONNECT,
2919 "write header to: %s failed: %E", http->hostport);
2921 rsp = error_response(csp, "connect-failed", errno);
2924 send_crunch_response(csp, rsp);
2934 * We're running an SSL tunnel and we're not forwarding,
2935 * so just send the "connect succeeded" message to the
2936 * client, flush the rest, and get out of the way.
2938 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2946 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2948 /* we're finished with the client's header */
2951 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2953 /* pass data between the client and server
2954 * until one or the other shuts down the connection.
2963 * FD_ZERO here seems to point to an errant macro which crashes.
2964 * So do this by hand for now...
2966 memset(&rfds,0x00,sizeof(fd_set));
2970 FD_SET(csp->cfd, &rfds);
2971 FD_SET(csp->sfd, &rfds);
2973 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2974 if ((csp->flags & CSP_FLAG_CHUNKED)
2975 && !(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
2976 && ((csp->iob->eod - csp->iob->cur) >= 5)
2977 && !memcmp(csp->iob->eod-5, "0\r\n\r\n", 5))
2979 log_error(LOG_LEVEL_CONNECT,
2980 "Looks like we read the last chunk together with "
2981 "the server headers. We better stop reading.");
2982 byte_count = (unsigned long long)(csp->iob->eod - csp->iob->cur);
2983 csp->expected_content_length = byte_count;
2984 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2986 if (server_body && server_response_is_complete(csp, byte_count))
2988 log_error(LOG_LEVEL_CONNECT,
2989 "Done reading from server. Expected content length: %llu. "
2990 "Actual content length: %llu. Most recently received: %d.",
2991 csp->expected_content_length, byte_count, len);
2994 * XXX: should not jump around,
2995 * chat() is complicated enough already.
2999 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
3001 timeout.tv_sec = csp->config->socket_timeout;
3002 timeout.tv_usec = 0;
3003 n = select((int)maxfd+1, &rfds, NULL, NULL, &timeout);
3007 log_error(LOG_LEVEL_ERROR,
3008 "Didn't receive data in time: %s", http->url);
3009 if ((byte_count == 0) && (http->ssl == 0))
3011 write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE,
3012 strlen(CONNECTION_TIMEOUT_RESPONSE));
3014 mark_server_socket_tainted(csp);
3019 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
3020 mark_server_socket_tainted(csp);
3025 * This is the body of the browser's request,
3026 * just read and write it.
3028 * XXX: Make sure the client doesn't use pipelining
3029 * behind Privoxy's back.
3031 if (FD_ISSET(csp->cfd, &rfds))
3033 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
3037 /* XXX: not sure if this is necessary. */
3038 mark_server_socket_tainted(csp);
3039 break; /* "game over, man" */
3042 if (write_socket(csp->sfd, buf, (size_t)len))
3044 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
3045 mark_server_socket_tainted(csp);
3052 * The server wants to talk. It could be the header or the body.
3053 * If `hdr' is null, then it's the header otherwise it's the body.
3054 * FIXME: Does `hdr' really mean `host'? No.
3056 if (FD_ISSET(csp->sfd, &rfds))
3059 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
3063 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
3065 if (http->ssl && (fwd->forward_host == NULL))
3068 * Just hang up. We already confirmed the client's CONNECT
3069 * request with status code 200 and unencrypted content is
3070 * no longer welcome.
3072 log_error(LOG_LEVEL_ERROR,
3073 "CONNECT already confirmed. Unable to tell the client about the problem.");
3076 else if (byte_count)
3079 * Just hang up. We already transmitted the original headers
3080 * and parts of the original content and therefore missed the
3081 * chance to send an error message (without risking data corruption).
3083 * XXX: we could retry with a fancy range request here.
3085 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
3086 "Unable to tell the client about the problem.");
3087 mark_server_socket_tainted(csp);
3091 * XXX: Consider handling the cases above the same.
3093 mark_server_socket_tainted(csp);
3097 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3098 if (csp->flags & CSP_FLAG_CHUNKED)
3100 if ((len >= 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
3102 /* XXX: this is a temporary hack */
3103 log_error(LOG_LEVEL_CONNECT,
3104 "Looks like we reached the end of the last chunk. "
3105 "We better stop reading.");
3106 csp->expected_content_length = byte_count + (unsigned long long)len;
3107 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
3111 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
3114 * Add a trailing zero to let be able to use string operations.
3115 * XXX: do we still need this with filter_popups gone?
3120 * Normally, this would indicate that we've read
3121 * as much as the server has sent us and we can
3122 * close the client connection. However, Microsoft
3123 * in its wisdom has released IIS/5 with a bug that
3124 * prevents it from sending the trailing \r\n in
3125 * a 302 redirect header (and possibly other headers).
3126 * To work around this if we've haven't parsed
3127 * a full header we'll append a trailing \r\n
3128 * and see if this now generates a valid one.
3130 * This hack shouldn't have any impacts. If we've
3131 * already transmitted the header or if this is a
3132 * SSL connection, then we won't bother with this
3133 * hack. So we only work on partially received
3134 * headers. If we append a \r\n and this still
3135 * doesn't generate a valid header, then we won't
3136 * transmit anything to the client.
3141 if (server_body || http->ssl)
3144 * If we have been buffering up the document,
3145 * now is the time to apply content modification
3146 * and send the result to the client.
3150 p = execute_content_filter(csp, content_filter);
3152 * If the content filter fails, use the original
3153 * buffer and length.
3154 * (see p != NULL ? p : csp->iob->cur below)
3158 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
3161 if (JB_ERR_OK != update_server_headers(csp))
3163 log_error(LOG_LEVEL_FATAL,
3164 "Failed to update server headers. after filtering.");
3167 hdr = list_to_text(csp->headers);
3170 /* FIXME Should handle error properly */
3171 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3174 if (write_socket(csp->cfd, hdr, strlen(hdr))
3175 || write_socket(csp->cfd,
3176 ((p != NULL) ? p : csp->iob->cur), (size_t)csp->content_length))
3178 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
3181 mark_server_socket_tainted(csp);
3189 break; /* "game over, man" */
3193 * This is NOT the body, so
3194 * Let's pretend the server just sent us a blank line.
3196 snprintf(buf, sizeof(buf), "\r\n");
3197 len = (int)strlen(buf);
3200 * Now, let the normal header parsing algorithm below do its
3201 * job. If it fails, we'll exit instead of continuing.
3208 * If this is an SSL connection or we're in the body
3209 * of the server document, just write it to the client,
3210 * unless we need to buffer the body for later content-filtering
3212 if (server_body || http->ssl)
3217 * If there is no memory left for buffering the content, or the buffer limit
3218 * has been reached, switch to non-filtering mode, i.e. make & write the
3219 * header, flush the iob and buf, and get out of the way.
3221 if (add_to_iob(csp, buf, len))
3226 log_error(LOG_LEVEL_INFO,
3227 "Flushing header and buffers. Stepping back from filtering.");
3229 hdr = list_to_text(csp->headers);
3233 * Memory is too tight to even generate the header.
3234 * Send our static "Out-of-memory" page.
3236 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
3237 rsp = cgi_error_memory();
3238 send_crunch_response(csp, rsp);
3239 mark_server_socket_tainted(csp);
3242 hdrlen = strlen(hdr);
3244 if (write_socket(csp->cfd, hdr, hdrlen)
3245 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
3246 || (write_socket(csp->cfd, buf, (size_t)len)))
3248 log_error(LOG_LEVEL_CONNECT,
3249 "Flush header and buffers to client failed: %E");
3251 mark_server_socket_tainted(csp);
3256 * Reset the byte_count to the amount of bytes
3257 * we just flushed. len will be added a few lines below,
3258 * hdrlen doesn't matter for LOG_LEVEL_CLF.
3260 byte_count = (unsigned long long)flushed;
3262 content_filter = NULL;
3268 if (write_socket(csp->cfd, buf, (size_t)len))
3270 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
3271 mark_server_socket_tainted(csp);
3275 byte_count += (unsigned long long)len;
3280 const char *header_start;
3282 * We're still looking for the end of the server's header.
3283 * Buffer up the data we just read. If that fails, there's
3284 * little we can do but send our static out-of-memory page.
3286 if (add_to_iob(csp, buf, len))
3288 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
3289 rsp = cgi_error_memory();
3290 send_crunch_response(csp, rsp);
3291 mark_server_socket_tainted(csp);
3295 header_start = csp->iob->cur;
3297 /* Convert iob into something sed() can digest */
3298 if (JB_ERR_PARSE == get_server_headers(csp))
3303 * Well, we tried our MS IIS/5 hack and it didn't work.
3304 * The header is incomplete and there isn't anything
3305 * we can do about it.
3307 log_error(LOG_LEVEL_ERROR, "Invalid server headers. "
3308 "Applying the MS IIS5 hack didn't help.");
3309 log_error(LOG_LEVEL_CLF,
3310 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3311 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3312 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3313 mark_server_socket_tainted(csp);
3319 * Since we have to wait for more from the server before
3320 * we can parse the headers we just continue here.
3322 long header_offset = csp->iob->cur - header_start;
3323 assert(csp->iob->cur >= header_start);
3324 byte_count += (unsigned long long)(len - header_offset);
3325 log_error(LOG_LEVEL_CONNECT, "Continuing buffering headers. "
3326 "byte_count: %llu. header_offset: %d. len: %d.",
3327 byte_count, header_offset, len);
3332 /* Did we actually get anything? */
3333 if (NULL == csp->headers->first)
3335 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
3336 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3337 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
3338 free_http_request(http);
3339 mark_server_socket_tainted(csp);
3343 assert(csp->headers->first->str);
3345 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
3346 strncmpic(csp->headers->first->str, "ICY", 3))
3349 * It doesn't look like a HTTP (or Shoutcast) response:
3350 * tell the client and log the problem.
3352 if (strlen(csp->headers->first->str) > 30)
3354 csp->headers->first->str[30] = '\0';
3356 log_error(LOG_LEVEL_ERROR,
3357 "Invalid server or forwarder response. Starts with: %s",
3358 csp->headers->first->str);
3359 log_error(LOG_LEVEL_CLF,
3360 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3361 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3362 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3363 free_http_request(http);
3364 mark_server_socket_tainted(csp);
3369 * We have now received the entire server header,
3370 * filter it and send the result to the client
3372 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
3374 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
3376 hdr = list_to_text(csp->headers);
3379 /* FIXME Should handle error properly */
3380 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3383 if (crunch_response_triggered(csp, crunchers_light))
3386 * One of the tags created by a server-header
3387 * tagger triggered a crunch. We already
3388 * delivered the crunch response to the client
3389 * and are done here after cleaning up.
3392 mark_server_socket_tainted(csp);
3395 /* Buffer and pcrs filter this if appropriate. */
3397 if (!http->ssl) /* We talk plaintext */
3399 content_filter = get_filter_function(csp);
3402 * Only write if we're not buffering for content modification
3404 if (!content_filter)
3407 * Write the server's (modified) header to
3408 * the client (along with anything else that
3409 * may be in the buffer)
3412 if (write_socket(csp->cfd, hdr, strlen(hdr))
3413 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
3415 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
3418 * The write failed, so don't bother mentioning it
3419 * to the client... it probably can't hear us anyway.
3422 mark_server_socket_tainted(csp);
3426 byte_count += (unsigned long long)len;
3431 * XXX: the header lenght should probably
3432 * be calculated by get_server_headers().
3434 long header_length = csp->iob->cur - header_start;
3435 assert(csp->iob->cur > header_start);
3436 byte_count += (unsigned long long)(len - header_length);
3439 /* we're finished with the server's header */
3445 * If this was a MS IIS/5 hack then it means the server
3446 * has already closed the connection. Nothing more to read.
3451 log_error(LOG_LEVEL_ERROR,
3452 "Closed server connection detected. "
3453 "Applying the MS IIS5 hack didn't help.");
3454 log_error(LOG_LEVEL_CLF,
3455 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3456 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3457 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3458 mark_server_socket_tainted(csp);
3464 mark_server_socket_tainted(csp);
3465 return; /* huh? we should never get here */
3468 if (csp->content_length == 0)
3471 * If Privoxy didn't recalculate the Content-Lenght,
3472 * byte_count is still correct.
3474 csp->content_length = byte_count;
3477 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3478 if ((csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
3479 && (csp->expected_content_length != byte_count))
3481 log_error(LOG_LEVEL_CONNECT,
3482 "Received %llu bytes while expecting %llu.",
3483 byte_count, csp->expected_content_length);
3484 mark_server_socket_tainted(csp);
3488 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %llu",
3489 csp->ip_addr_str, http->ocmd, csp->content_length);
3493 /*********************************************************************
3497 * Description : This is little more than chat. We only "serve" to
3498 * to close (or remember) any socket that chat may have
3502 * 1 : csp = Current client state (buffers, headers, etc...)
3506 *********************************************************************/
3508 void serve(struct client_state *csp)
3509 #else /* ifndef AMIGA */
3510 static void serve(struct client_state *csp)
3511 #endif /* def AMIGA */
3513 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3514 static int monitor_thread_running = 0;
3515 int continue_chatting = 0;
3520 continue_chatting = (csp->config->feature_flags
3521 & RUNTIME_FEATURE_CONNECTION_KEEP_ALIVE)
3522 && (csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE)
3523 && (csp->cfd != JB_INVALID_SOCKET)
3524 && (csp->sfd != JB_INVALID_SOCKET)
3525 && socket_is_still_usable(csp->sfd);
3527 if (continue_chatting)
3529 log_error(LOG_LEVEL_CONNECT,
3530 "Waiting for the next client request. "
3531 "Keeping the server socket %d to %s open.",
3532 csp->sfd, csp->server_connection.host);
3534 if ((csp->flags & CSP_FLAG_CLIENT_CONNECTION_KEEP_ALIVE)
3535 && data_is_available(csp->cfd, csp->config->keep_alive_timeout)
3536 && socket_is_still_usable(csp->cfd))
3538 log_error(LOG_LEVEL_CONNECT, "Client request arrived in "
3539 "time or the client closed the connection.");
3541 * Get the csp in a mostly vergin state again.
3542 * XXX: Should be done elsewhere.
3544 csp->content_type = 0;
3545 csp->content_length = 0;
3546 csp->expected_content_length = 0;
3547 list_remove_all(csp->headers);
3548 freez(csp->iob->buf);
3549 memset(csp->iob, 0, sizeof(csp->iob));
3550 freez(csp->error_message);
3551 free_http_request(csp->http);
3552 destroy_list(csp->headers);
3553 destroy_list(csp->tags);
3554 free_current_action(csp->action);
3555 if (NULL != csp->fwd)
3557 unload_forward_spec(csp->fwd);
3561 /* XXX: Store per-connection flags someplace else. */
3562 csp->flags = CSP_FLAG_ACTIVE | (csp->flags & CSP_FLAG_TOGGLED_ON);
3566 log_error(LOG_LEVEL_CONNECT,
3567 "No additional client request received in time.");
3568 if ((csp->config->feature_flags & RUNTIME_FEATURE_CONNECTION_SHARING))
3570 remember_connection(csp->sfd, csp->http,
3571 forward_url(csp, csp->http));
3572 csp->sfd = JB_INVALID_SOCKET;
3573 close_socket(csp->cfd);
3574 csp->cfd = JB_INVALID_SOCKET;
3575 privoxy_mutex_lock(&connection_reuse_mutex);
3576 if (!monitor_thread_running)
3578 monitor_thread_running = 1;
3579 privoxy_mutex_unlock(&connection_reuse_mutex);
3580 wait_for_alive_connections();
3581 privoxy_mutex_lock(&connection_reuse_mutex);
3582 monitor_thread_running = 0;
3584 privoxy_mutex_unlock(&connection_reuse_mutex);
3589 else if (csp->sfd != JB_INVALID_SOCKET)
3591 log_error(LOG_LEVEL_CONNECT,
3592 "The connection on server socket %d to %s isn't reusable. "
3593 "Closing.", csp->sfd, csp->server_connection.host);
3595 } while (continue_chatting);
3597 mark_connection_closed(&csp->server_connection);
3600 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3602 if (csp->sfd != JB_INVALID_SOCKET)
3604 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3605 forget_connection(csp->sfd);
3606 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3607 close_socket(csp->sfd);
3610 if (csp->cfd != JB_INVALID_SOCKET)
3612 close_socket(csp->cfd);
3615 csp->flags &= ~CSP_FLAG_ACTIVE;
3621 /*********************************************************************
3623 * Function : server_thread
3625 * Description : We only exist to call `serve' in a threaded environment.
3628 * 1 : data = Current client state (buffers, headers, etc...)
3630 * Returns : Always 0.
3632 *********************************************************************/
3633 static int32 server_thread(void *data)
3635 serve((struct client_state *) data);
3642 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3643 /*********************************************************************
3647 * Description : Print usage info & exit.
3649 * Parameters : Pointer to argv[0] for identifying ourselves
3653 *********************************************************************/
3654 static void usage(const char *myname)
3656 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3660 #endif /* defined(unix) */
3663 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3664 #endif /* defined(unix) */
3665 "[--version] [configfile]\n"
3666 "Aborting\n", myname);
3671 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3674 #ifdef MUTEX_LOCKS_AVAILABLE
3675 /*********************************************************************
3677 * Function : privoxy_mutex_lock
3679 * Description : Locks a mutex.
3682 * 1 : mutex = The mutex to lock.
3684 * Returns : Void. May exit in case of errors.
3686 *********************************************************************/
3687 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3689 #ifdef FEATURE_PTHREAD
3690 int err = pthread_mutex_lock(mutex);
3693 if (mutex != &log_mutex)
3695 log_error(LOG_LEVEL_FATAL,
3696 "Mutex locking failed: %s.\n", strerror(err));
3701 EnterCriticalSection(mutex);
3702 #endif /* def FEATURE_PTHREAD */
3706 /*********************************************************************
3708 * Function : privoxy_mutex_unlock
3710 * Description : Unlocks a mutex.
3713 * 1 : mutex = The mutex to unlock.
3715 * Returns : Void. May exit in case of errors.
3717 *********************************************************************/
3718 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3720 #ifdef FEATURE_PTHREAD
3721 int err = pthread_mutex_unlock(mutex);
3724 if (mutex != &log_mutex)
3726 log_error(LOG_LEVEL_FATAL,
3727 "Mutex unlocking failed: %s.\n", strerror(err));
3732 LeaveCriticalSection(mutex);
3733 #endif /* def FEATURE_PTHREAD */
3737 /*********************************************************************
3739 * Function : privoxy_mutex_init
3741 * Description : Prepares a mutex.
3744 * 1 : mutex = The mutex to initialize.
3746 * Returns : Void. May exit in case of errors.
3748 *********************************************************************/
3749 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3751 #ifdef FEATURE_PTHREAD
3752 int err = pthread_mutex_init(mutex, 0);
3755 printf("Fatal error. Mutex initialization failed: %s.\n",
3760 InitializeCriticalSection(mutex);
3761 #endif /* def FEATURE_PTHREAD */
3763 #endif /* def MUTEX_LOCKS_AVAILABLE */
3765 /*********************************************************************
3767 * Function : initialize_mutexes
3769 * Description : Prepares mutexes if mutex support is available.
3773 * Returns : Void, exits in case of errors.
3775 *********************************************************************/
3776 static void initialize_mutexes(void)
3778 #ifdef MUTEX_LOCKS_AVAILABLE
3780 * Prepare global mutex semaphores
3782 privoxy_mutex_init(&log_mutex);
3783 privoxy_mutex_init(&log_init_mutex);
3784 privoxy_mutex_init(&connection_reuse_mutex);
3787 * XXX: The assumptions below are a bit naive
3788 * and can cause locks that aren't necessary.
3790 * For example older FreeBSD versions (< 6.x?)
3791 * have no gethostbyname_r, but gethostbyname is
3794 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3795 privoxy_mutex_init(&resolver_mutex);
3796 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3798 * XXX: should we use a single mutex for
3799 * localtime() and gmtime() as well?
3801 #ifndef HAVE_GMTIME_R
3802 privoxy_mutex_init(&gmtime_mutex);
3803 #endif /* ndef HAVE_GMTIME_R */
3805 #ifndef HAVE_LOCALTIME_R
3806 privoxy_mutex_init(&localtime_mutex);
3807 #endif /* ndef HAVE_GMTIME_R */
3810 privoxy_mutex_init(&rand_mutex);
3811 #endif /* ndef HAVE_RANDOM */
3812 #endif /* def MUTEX_LOCKS_AVAILABLE */
3816 /*********************************************************************
3820 * Description : Load the config file and start the listen loop.
3821 * This function is a lot more *sane* with the `load_config'
3822 * and `listen_loop' functions; although it stills does
3823 * a *little* too much for my taste.
3826 * 1 : argc = Number of parameters (including $0).
3827 * 2 : argv = Array of (char *)'s to the parameters.
3829 * Returns : 1 if : can't open config file, unrecognized directive,
3830 * stats requested in multi-thread mode, can't open the
3831 * log file, can't open the jar file, listen port is invalid,
3832 * any load fails, and can't bind port.
3834 * Else main never returns, the process must be signaled
3835 * to terminate execution. Or, on Windows, use the
3836 * "File", "Exit" menu option.
3838 *********************************************************************/
3840 int real_main(int argc, const char *argv[])
3842 int main(int argc, const char *argv[])
3846 unsigned int random_seed;
3848 struct passwd *pw = NULL;
3849 struct group *grp = NULL;
3852 char *pre_chroot_nslookup_to_load_resolver = NULL;
3859 #if !defined(_WIN32)
3866 /* Prepare mutexes if supported and necessary. */
3867 initialize_mutexes();
3869 /* Enable logging until further notice. */
3873 * Parse the command line arguments
3875 * XXX: simply printing usage information in case of
3876 * invalid arguments isn't particularly user friendly.
3878 while (++argc_pos < argc)
3881 /* Check to see if the service must be installed or uninstalled */
3882 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3884 const char *pName = argv[argc_pos] + 9;
3887 exit( (install_service(pName)) ? 0 : 1 );
3889 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3891 const char *pName = argv[argc_pos] + 11;
3894 exit((uninstall_service(pName)) ? 0 : 1);
3896 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3898 bRunAsService = TRUE;
3899 w32_set_service_cwd();
3900 atexit(w32_service_exit_notify);
3903 #endif /* defined(_WIN32) */
3906 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3908 if (strcmp(argv[argc_pos], "--help") == 0)
3913 else if(strcmp(argv[argc_pos], "--version") == 0)
3915 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3921 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3923 set_debug_level(LOG_LEVEL_FATAL | LOG_LEVEL_ERROR | LOG_LEVEL_INFO);
3927 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3929 if (++argc_pos == argc) usage(argv[0]);
3930 pidfile = strdup(argv[argc_pos]);
3933 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3935 if (++argc_pos == argc) usage(argv[argc_pos]);
3937 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3940 if (NULL == (grp = getgrnam(p)))
3942 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3946 if (NULL == (pw = getpwnam(argv[argc_pos])))
3948 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3951 if (p != NULL) *--p = '\0';
3954 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3956 if (++argc_pos == argc) usage(argv[0]);
3957 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3960 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3964 #endif /* defined(unix) */
3966 else if (argc_pos + 1 != argc)
3969 * This is neither the last command line
3970 * option, nor was it recognized before,
3971 * therefore it must be invalid.
3977 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3979 configfile = argv[argc_pos];
3982 } /* -END- while (more arguments) */
3984 show_version(Argv[0]);
3987 if ( *configfile != '/' )
3989 char cwd[BUFFER_SIZE];
3991 size_t abs_file_size;
3993 /* make config-filename absolute here */
3994 if (NULL == getcwd(cwd, sizeof(cwd)))
3996 perror("failed to get current working directory");
4001 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
4002 basedir = strdup(cwd);
4004 if (NULL == basedir ||
4005 NULL == (abs_file = malloc(abs_file_size)))
4007 perror("malloc failed");
4010 strlcpy(abs_file, basedir, abs_file_size);
4011 strlcat(abs_file, "/", abs_file_size );
4012 strlcat(abs_file, configfile, abs_file_size);
4013 configfile = abs_file;
4015 #endif /* defined unix */
4019 clients->next = NULL;
4021 /* XXX: factor out initialising after the next stable release. */
4024 #elif defined(_WIN32)
4028 random_seed = (unsigned int)time(NULL);
4030 srandom(random_seed);
4033 #endif /* ifdef HAVE_RANDOM */
4036 * Unix signal handling
4038 * Catch the abort, interrupt and terminate signals for a graceful exit
4039 * Catch the hangup signal so the errlog can be reopened.
4040 * Ignore the broken pipe signals (FIXME: Why?)
4042 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
4045 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
4046 const int ignored_signals[] = { SIGPIPE, 0 };
4048 for (idx = 0; catched_signals[idx] != 0; idx++)
4050 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
4051 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
4053 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
4054 #endif /* ifdef sun */
4056 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
4060 for (idx = 0; ignored_signals[idx] != 0; idx++)
4062 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
4064 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
4069 #else /* ifdef _WIN32 */
4070 # ifdef _WIN_CONSOLE
4072 * We *are* in a windows console app.
4073 * Print a verbose messages about FAQ's and such
4075 printf("%s", win32_blurb);
4076 # endif /* def _WIN_CONSOLE */
4077 #endif /* def _WIN32 */
4080 /* Initialize the CGI subsystem */
4081 cgi_init_error_messages();
4084 * If runnig on unix and without the --nodaemon
4085 * option, become a daemon. I.e. fork, detach
4086 * from tty and get process group leadership
4099 if ( pid < 0 ) /* error */
4104 else if ( pid != 0 ) /* parent */
4109 * must check for errors
4110 * child died due to missing files aso
4113 wpid = waitpid( pid, &status, WNOHANG );
4122 /* Should be more portable, but not as well tested */
4127 #else /* ndef __FreeBSD__ */
4129 #endif /* ndef __FreeBSD__ */
4130 fd = open("/dev/tty", O_RDONLY);
4133 /* no error check here */
4134 ioctl( fd, TIOCNOTTY,0 );
4139 * stderr (fd 2) will be closed later on,
4140 * when the config file has been parsed.
4147 } /* -END- if (!no_daemon) */
4150 * As soon as we have written the PID file, we can switch
4151 * to the user and group ID indicated by the --user option
4157 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
4159 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
4163 if (setgroups(1, &grp->gr_gid))
4165 log_error(LOG_LEVEL_FATAL, "setgroups() failed: %E");
4168 else if (initgroups(pw->pw_name, pw->pw_gid))
4170 log_error(LOG_LEVEL_FATAL, "initgroups() failed: %E");
4176 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
4178 /* Read the time zone file from /etc before doing chroot. */
4180 if (NULL != pre_chroot_nslookup_to_load_resolver
4181 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
4183 /* Initialize resolver library. */
4184 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
4186 if (chroot(pw->pw_dir) < 0)
4188 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
4192 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
4195 if (setuid(pw->pw_uid))
4197 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
4201 char putenv_dummy[64];
4203 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
4204 if (putenv(putenv_dummy) != 0)
4206 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
4209 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
4210 if (putenv(putenv_dummy) != 0)
4212 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
4218 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
4221 #endif /* defined unix */
4224 /* This will be FALSE unless the command line specified --service
4228 /* Yup, so now we must attempt to establish a connection
4229 * with the service dispatcher. This will only work if this
4230 * process was launched by the service control manager to
4231 * actually run as a service. If this isn't the case, i've
4232 * known it take around 30 seconds or so for the call to return.
4235 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
4236 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
4238 /* Service has run, and at this point is now being stopped, so just return */
4243 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
4245 /* An error occurred. Usually it's because --service was wrongly specified
4246 * and we were unable to connect to the Service Control Dispatcher because
4247 * it wasn't expecting us and is therefore not listening.
4249 * For now, just continue below to call the listen_loop function.
4252 #endif /* def _WIN32 */
4262 /*********************************************************************
4264 * Function : bind_port_helper
4266 * Description : Bind the listen port. Handles logging, and aborts
4270 * 1 : config = Privoxy configuration. Specifies port
4273 * Returns : Port that was opened.
4275 *********************************************************************/
4276 static jb_socket bind_port_helper(struct configuration_spec * config)
4281 if (config->haddr == NULL)
4283 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
4288 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
4289 config->hport, config->haddr);
4292 result = bind_port(config->haddr, config->hport, &bfd);
4299 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
4300 "There may be another Privoxy or some other "
4301 "proxy running on port %d",
4302 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
4303 config->hport, config->hport);
4306 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
4307 "The hostname is not resolvable",
4308 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
4311 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
4312 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
4315 /* shouldn't get here */
4316 return JB_INVALID_SOCKET;
4319 config->need_bind = 0;
4326 /* Without this simple workaround we get this compiler warning from _beginthread
4327 * warning C4028: formal parameter 1 different from declaration
4329 void w32_service_listen_loop(void *p)
4333 #endif /* def _WIN32 */
4336 /*********************************************************************
4338 * Function : listen_loop
4340 * Description : bind the listen port and enter a "FOREVER" listening loop.
4346 *********************************************************************/
4347 static void listen_loop(void)
4349 struct client_state *csp = NULL;
4351 struct configuration_spec *config;
4352 unsigned int active_threads = 0;
4354 config = load_config();
4356 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
4358 * XXX: Should be relocated once it no
4359 * longer needs to emit log messages.
4361 initialize_reusable_connections();
4362 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
4364 bfd = bind_port_helper(config);
4366 #ifdef FEATURE_GRACEFUL_TERMINATION
4367 while (!g_terminate)
4372 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
4373 while (waitpid(-1, NULL, WNOHANG) > 0)
4375 /* zombie children */
4377 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
4380 * Free data that was used by died threads
4382 active_threads = sweep();
4386 * Re-open the errlog after HUP signal
4388 if (received_hup_signal)
4390 if (NULL != config->logfile)
4392 init_error_log(Argv[0], config->logfile);
4394 received_hup_signal = 0;
4398 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
4400 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
4404 csp->flags |= CSP_FLAG_ACTIVE;
4405 csp->sfd = JB_INVALID_SOCKET;
4407 csp->config = config = load_config();
4409 if ( config->need_bind )
4412 * Since we were listening to the "old port", we will not see
4413 * a "listen" param change until the next request. So, at
4414 * least 1 more request must be made for us to find the new
4415 * setting. I am simply closing the old socket and binding the
4418 * Which-ever is correct, we will serve 1 more page via the
4419 * old settings. This should probably be a "show-proxy-args"
4420 * request. This should not be a so common of an operation
4421 * that this will hurt people's feelings.
4426 bfd = bind_port_helper(config);
4429 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
4431 if (!accept_connection(csp, bfd))
4433 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
4446 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
4449 #ifdef FEATURE_TOGGLE
4450 if (global_toggle_state)
4451 #endif /* def FEATURE_TOGGLE */
4453 csp->flags |= CSP_FLAG_TOGGLED_ON;
4456 if (run_loader(csp))
4458 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
4459 /* Never get here - LOG_LEVEL_FATAL causes program exit */
4463 if (block_acl(NULL,csp))
4465 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
4466 close_socket(csp->cfd);
4467 freez(csp->ip_addr_str);
4471 #endif /* def FEATURE_ACL */
4473 if ((0 != config->max_client_connections)
4474 && (active_threads >= config->max_client_connections))
4476 log_error(LOG_LEVEL_CONNECT,
4477 "Rejecting connection from %s. Maximum number of connections reached.",
4479 write_socket(csp->cfd, TOO_MANY_CONNECTIONS_RESPONSE,
4480 strlen(TOO_MANY_CONNECTIONS_RESPONSE));
4481 close_socket(csp->cfd);
4482 freez(csp->ip_addr_str);
4487 /* add it to the list of clients */
4488 csp->next = clients->next;
4489 clients->next = csp;
4491 if (config->multi_threaded)
4495 /* this is a switch () statment in the C preprocessor - ugh */
4496 #undef SELECTED_ONE_OPTION
4498 /* Use Pthreads in preference to native code */
4499 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
4500 #define SELECTED_ONE_OPTION
4502 pthread_t the_thread;
4503 pthread_attr_t attrs;
4505 pthread_attr_init(&attrs);
4506 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
4507 errno = pthread_create(&the_thread, &attrs,
4508 (void * (*)(void *))serve, csp);
4509 child_id = errno ? -1 : 0;
4510 pthread_attr_destroy(&attrs);
4514 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
4515 #define SELECTED_ONE_OPTION
4516 child_id = _beginthread(
4517 (void (*)(void *))serve,
4522 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
4523 #define SELECTED_ONE_OPTION
4524 child_id = _beginthread(
4525 (void(* _Optlink)(void*))serve,
4531 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
4532 #define SELECTED_ONE_OPTION
4534 thread_id tid = spawn_thread
4535 (server_thread, "server", B_NORMAL_PRIORITY, csp);
4537 if ((tid >= 0) && (resume_thread(tid) == B_OK))
4539 child_id = (int) tid;
4548 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
4549 #define SELECTED_ONE_OPTION
4550 csp->cfd = ReleaseSocket(csp->cfd, -1);
4553 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
4554 NP_Output, Output(),
4555 NP_CloseOutput, FALSE,
4556 NP_Name, (ULONG)"privoxy child",
4560 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
4561 NP_Output, Output(),
4562 NP_CloseOutput, FALSE,
4563 NP_Name, (ULONG)"privoxy child",
4564 NP_StackSize, 200*1024,
4570 ((struct Task *)child_id)->tc_UserData = csp;
4571 Signal((struct Task *)child_id, SIGF_SINGLE);
4576 #if !defined(SELECTED_ONE_OPTION)
4579 /* This block is only needed when using fork().
4580 * When using threads, the server thread was
4581 * created and run by the call to _beginthread().
4583 if (child_id == 0) /* child */
4586 #ifdef FEATURE_TOGGLE
4587 int inherited_toggle_state = global_toggle_state;
4588 #endif /* def FEATURE_TOGGLE */
4593 * If we've been toggled or we've blocked the request, tell Mom
4596 #ifdef FEATURE_TOGGLE
4597 if (inherited_toggle_state != global_toggle_state)
4599 rc |= RC_FLAG_TOGGLED;
4601 #endif /* def FEATURE_TOGGLE */
4603 #ifdef FEATURE_STATISTICS
4604 if (csp->flags & CSP_FLAG_REJECTED)
4606 rc |= RC_FLAG_BLOCKED;
4608 #endif /* ndef FEATURE_STATISTICS */
4612 else if (child_id > 0) /* parent */
4614 /* in a fork()'d environment, the parent's
4615 * copy of the client socket and the CSP
4619 #if !defined(_WIN32) && !defined(__CYGWIN__)
4621 wait( &child_status );
4624 * Evaluate child's return code: If the child has
4625 * - been toggled, toggle ourselves
4626 * - blocked its request, bump up the stats counter
4629 #ifdef FEATURE_TOGGLE
4630 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
4632 global_toggle_state = !global_toggle_state;
4634 #endif /* def FEATURE_TOGGLE */
4636 #ifdef FEATURE_STATISTICS
4638 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4642 #endif /* def FEATURE_STATISTICS */
4644 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4645 close_socket(csp->cfd);
4646 csp->flags &= ~CSP_FLAG_ACTIVE;
4650 #undef SELECTED_ONE_OPTION
4651 /* end of cpp switch () */
4656 * Spawning the child failed, assume it's because
4657 * there are too many children running already.
4658 * XXX: If you assume ...
4660 log_error(LOG_LEVEL_ERROR,
4661 "Unable to take any additional connections: %E");
4662 write_socket(csp->cfd, TOO_MANY_CONNECTIONS_RESPONSE,
4663 strlen(TOO_MANY_CONNECTIONS_RESPONSE));
4664 close_socket(csp->cfd);
4665 csp->flags &= ~CSP_FLAG_ACTIVE;
4674 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4676 /* Clean up. Aim: free all memory (no leaks) */
4677 #ifdef FEATURE_GRACEFUL_TERMINATION
4679 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4681 unload_current_config_file();
4682 unload_current_actions_file();
4683 unload_current_re_filterfile();
4684 #ifdef FEATURE_TRUST
4685 unload_current_trust_file();
4688 if (config->multi_threaded)
4695 } while ((clients->next != NULL) && (--i > 0));
4699 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4710 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4711 /* Cleanup - remove taskbar icon etc. */
4716 #endif /* FEATURE_GRACEFUL_TERMINATION */