1 const char jcc_rcs[] = "$Id: jcc.c,v 1.235 2009/03/18 21:01:20 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 1.235 2009/03/18 21:01:20 fabiankeil
37 * Comment fix. Spotted by Roland.
39 * Revision 1.234 2009/03/18 20:48:42 fabiankeil
40 * If the --no-daemon option is used, enable LOG_LEVEL_INFO
41 * before the config file has been parsed (as we always did).
43 * Revision 1.233 2009/03/13 14:10:07 fabiankeil
44 * Fix some more harmless warnings on amd64.
46 * Revision 1.232 2009/03/08 19:29:16 fabiankeil
47 * Reinitialize the timeout structure every time before passing
48 * it to select(). Apparently some implementations mess with it.
49 * Probably fixes #2669131 reported by cyberpatrol.
51 * Revision 1.231 2009/03/08 14:19:23 fabiankeil
52 * Fix justified (but harmless) compiler warnings
53 * on platforms where sizeof(int) < sizeof(long).
55 * Revision 1.230 2009/03/07 13:09:17 fabiankeil
56 * Change csp->expected_content and_csp->expected_content_length from
57 * size_t to unsigned long long to reduce the likelihood of integer
58 * overflows that would let us close the connection prematurely.
59 * Bug found while investigating #2669131, reported by cyberpatrol.
61 * Revision 1.229 2009/03/07 11:17:01 fabiankeil
62 * Fix compiler warning.
64 * Revision 1.228 2009/03/06 20:30:13 fabiankeil
65 * Log unsigned values as such.
67 * Revision 1.227 2009/03/02 19:18:11 fabiankeil
68 * Streamline parse_http_request()'s prototype. As
69 * cparser pointed out it doesn't actually use csp.
71 * Revision 1.226 2009/03/01 18:28:24 fabiankeil
72 * Help clang understand that we aren't dereferencing
75 * Revision 1.225 2009/02/19 18:09:32 fabiankeil
76 * Unbreak build without FEATURE_CONNECTION_KEEP_ALIVE.
79 * Revision 1.224 2009/02/14 15:32:04 fabiankeil
80 * Add the request URL to the timeout message in chat().
83 * Revision 1.223 2009/02/09 21:21:16 fabiankeil
84 * Now that init_log_module() is called earlier, call show_version()
85 * later on from main() directly so it doesn't get called for --help
88 * Revision 1.222 2009/02/08 12:56:51 fabiankeil
89 * Call initialize_mutexes() before init_log_module() again.
90 * Broken since r220, might be the cause of Lee's #2579448.
92 * Revision 1.221 2009/02/06 18:02:58 fabiankeil
93 * When dropping privileges, also give up membership in supplementary
94 * groups. Thanks to Matthias Drochner for reporting the problem,
95 * providing the initial patch and testing the final version.
97 * Revision 1.220 2009/02/04 18:29:07 fabiankeil
98 * Initialize the log module before parsing arguments.
99 * Thanks to Matthias Drochner for the report.
101 * Revision 1.219 2009/01/31 16:08:21 fabiankeil
102 * Remove redundant error check in receive_client_request().
104 * Revision 1.218 2009/01/31 12:25:54 fabiankeil
105 * Flatten indentation in receive_client_request().
107 * Revision 1.217 2009/01/07 19:50:09 fabiankeil
108 * - If the socket-timeout has been reached and the client
109 * hasn't received any data yet, send an explanation before
110 * closing the connection.
111 * - In get_request_line(), signal timeouts the right way.
113 * Revision 1.216 2008/12/24 22:13:11 ler762
114 * fix GCC 3.4.4 warning
116 * Revision 1.215 2008/12/24 17:06:19 fabiankeil
117 * Keep a thread around to timeout alive connections
118 * even if no new requests are coming in.
120 * Revision 1.214 2008/12/20 14:53:55 fabiankeil
121 * Add config option socket-timeout to control the time
122 * Privoxy waits for data to arrive on a socket. Useful
123 * in case of stale ssh tunnels or when fuzz-testing.
125 * Revision 1.213 2008/12/15 18:45:51 fabiankeil
126 * When logging crunches, log the whole URL, so one can easily
127 * differentiate between vanilla HTTP and CONNECT requests.
129 * Revision 1.212 2008/12/14 15:46:22 fabiankeil
130 * Give crunched requests their own log level.
132 * Revision 1.211 2008/12/06 10:05:03 fabiankeil
133 * Downgrade "Received x bytes while expecting y." message to
134 * LOG_LEVEL_CONNECT as it doesn't necessarily indicate an error.
136 * Revision 1.210 2008/12/02 22:03:18 fabiankeil
137 * Don't miscalculate byte_count if we don't get all the
138 * server headers with one read_socket() call. With keep-alive
139 * support enabled, this caused delays until the server closed
142 * Revision 1.209 2008/11/27 09:44:04 fabiankeil
143 * Cosmetics for the last commit: Don't watch out for
144 * the last chunk if the content isn't chunk-encoded or
145 * if we already determined the content length previously.
147 * Revision 1.208 2008/11/26 18:24:17 fabiankeil
148 * Recognize that the server response is complete if the
149 * last chunk is read together with the server headers.
152 * Revision 1.207 2008/11/25 17:25:16 fabiankeil
153 * Don't convert the client-header list to text until we need to.
155 * Revision 1.206 2008/11/23 17:00:11 fabiankeil
156 * Some more chat() cosmetics.
158 * Revision 1.205 2008/11/16 12:43:49 fabiankeil
159 * Turn keep-alive support into a runtime feature
160 * that is disabled by setting keep-alive-timeout
161 * to a negative value.
163 * Revision 1.204 2008/11/06 19:42:17 fabiankeil
164 * Fix last-chunk detection hack to also apply
165 * if buf[] contains nothing but the last-chunk.
167 * Revision 1.203 2008/11/06 18:34:35 fabiankeil
168 * Factor receive_client_request() and
169 * parse_client_request() out of chat().
171 * Revision 1.202 2008/11/02 18:40:34 fabiankeil
172 * If we received a different amount of data than we expected,
173 * log a warning and make sure the server socket isn't reused.
175 * Revision 1.201 2008/11/02 16:48:20 fabiankeil
176 * Revert revision 1.195 and try again.
178 * Revision 1.200 2008/10/26 16:53:18 fabiankeil
181 * Revision 1.199 2008/10/26 15:36:10 fabiankeil
182 * Remove two debug messages with LOG_LEVEL_INFO.
184 * Revision 1.198 2008/10/22 15:19:55 fabiankeil
185 * Once More, With Feeling: if there is no logfile
186 * because the user didn't specify one, we shouldn't
187 * call init_error_log() after receiving SIGHUP either.
189 * Revision 1.197 2008/10/20 17:02:40 fabiankeil
190 * If SIGHUP is received while we aren't running in daemon
191 * mode, calling init_error_log() would be a mistake.
193 * Revision 1.196 2008/10/16 09:16:41 fabiankeil
194 * - Fix two gcc44 conversion warnings.
195 * - Don't bother logging the last five bytes
198 * Revision 1.195 2008/10/13 16:04:37 fabiankeil
199 * Make sure we don't try to reuse tainted server sockets.
201 * Revision 1.194 2008/10/12 18:35:18 fabiankeil
202 * The last commit was a bit too ambitious, apparently the content
203 * length adjustment is only necessary if we aren't buffering.
205 * Revision 1.193 2008/10/12 15:57:35 fabiankeil
206 * Fix content length calculation if we read headers
207 * and the start of the body at once. Now that we have
208 * FEATURE_CONNECTION_KEEP_ALIVE, it actually matters.
210 * Revision 1.192 2008/10/11 18:19:14 fabiankeil
211 * Even more chat() cosmetics.
213 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
214 * Reformat some comments in chat().
216 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
217 * In case of chunk-encoded content, stop reading if
218 * the buffer looks like it ends with the last chunk.
220 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
221 * Let server_response_is_complete() deal properly with
222 * content that is neither buffered nor read all at once.
224 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
225 * Flush work-in-progress changes to keep outgoing connections
226 * alive where possible. Incomplete and mostly #ifdef'd out.
228 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
229 * Add mutex lock support for _WIN32.
231 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
232 * Prepare for critical sections on Windows by adding a
233 * layer of indirection before the pthread mutex functions.
235 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
236 * Remove FEATURE_COOKIE_JAR.
238 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
239 * - Silence LLVM/Clang complaint.
240 * - Make received_hup_signal static.
241 * - Hide definitions for basedir, pidfile and received_hup_signal
242 * from __EMX__ as they only seem to be used in case of #ifdef unix.
244 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
245 * Accept Shoutcast responses again. Problem reported
246 * and fix suggested by Stefan in #2062860.
248 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
249 * Fix possible NULL-pointer dereference reported
250 * by din_a4 in #2003937. Pointy hat to me.
252 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
253 * Streamline sed()'s prototype and declare
254 * the header parse and add structures static.
256 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
257 * - Mark csp as immutable for send_crunch_response().
258 * - Fix comment spelling.
260 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
261 * Factor update_server_headers() out of sed(), ditch the
262 * first_run hack and make server_patterns_light static.
264 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
265 * Don't provide get_header() with the whole client state
266 * structure when it only needs access to csp->iob.
268 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
269 * Make the "read the rest of the headers" loop a bit more readable.
271 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
272 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
273 * - Remove useless comment.
275 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
276 * Fix comment grammar.
278 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
279 * Remove the pointless buffer in client_protocol_is_unsupported().
281 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
282 * Least-effort fix for bug #1821930 (reported by Lee):
283 * If the response doesn't look like HTTP,
284 * tell the client and log the problem.
286 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
287 * Don't pass the whole csp structure to flush_socket()
288 * when it only needs a file descriptor and a buffer.
290 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
291 * Remove kill-popups action.
293 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
294 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
296 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
297 * Remove the treat-forbidden-connects-like-blocks action. We now
298 * use the "blocked" page for forbidden CONNECT requests by default.
300 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
301 * Also use shiny new connect_port_is_forbidden() in jcc.c.
303 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
304 * Rename url_actions() to get_url_actions() and let it
305 * use the standard parameter ordering.
307 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
308 * Let forward_url() use the standard parameter ordering
309 * and mark its second parameter immutable.
311 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
312 * Remove the "Listening ... for local connections only" log message.
313 * Whether or not remote connections are able to reach Privoxy is up
314 * to the operating system.
316 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
317 * Prevent the log messages for CONNECT requests to unacceptable
318 * ports from printing the limit-connect argument as [null] if
319 * limit-connect hasn't been explicitly enabled.
321 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
322 * Make sure all console-mode apps get a usage() instance
324 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
325 * Reword NO_SERVER_DATA_RESPONSE to make it harder
326 * to misunderstand what the message is all about.
328 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
329 * Unbreak trustfile which previously didn't work without
330 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
332 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
333 * Plug memory leak. Spotted by Valgrind, triggered by
334 * Privoxy-Regression-Test feeding proxyfuzz.py.
336 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
337 * In the HTTP snipplets, refer to the client as client.
339 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
340 * Emit a log message when activating the MS IIS5 hack.
342 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
343 * Log the "weak randomization factor" warning only
344 * once for mingw32 and provide some more details.
346 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
347 * Initialize log module after initializing mutexes, future
348 * deadlocks in that code should now work cross-platform.
350 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
351 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
352 * Reported by Bert van Leeuwen in BR#1818808.
354 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
355 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
357 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
358 * When in daemon mode, close stderr after the configuration file has been
359 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
361 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
362 * - Fix a crash when parsing invalid requests whose first header
363 * is rejected by get_header(). Regression (re?)introduced
364 * in r1.143 by yours truly.
365 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
366 * client_cookie_adder() to make sure send-vanilla-wafer can be
367 * controlled through tags (and thus regression-tested).
369 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
370 * - Move get_filter_function() from jcc.c to filters.c
371 * so the filter functions can be static.
372 * - Don't bother filtering body-less responses.
374 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
375 * Execute content filters through execute_content_filter().
377 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
378 * Initialize req to NULL to make sure it's defined if the
379 * first read_socket() call fails. Reported by icmp30.
381 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
382 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
383 * extensive comments moved to user manual.
385 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
386 * Don't crash if a broken header filter wiped out the request line.
388 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
389 * Fix byte_count calculation in case of flushes
390 * and don't parse the server headers a second time.
392 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
393 * - If there's a connection problem after we already forwarded
394 * parts of the original content, just hang up. Fixes BR#1776724.
395 * - Fix warnings about unused code on mingw32.
396 * - In case of flushes, calculate the byte count
397 * less incorrectly (I think).
399 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
400 * Add some more prototypes for static functions.
402 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
403 * Comment out request_contains_null_bytes() until it's used again.
405 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
406 * #1763173 from Stefan Huehner: s@const static@static const@
407 * and declare some more functions static.
409 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
410 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
411 * and explain why they were rejected in the first place.
412 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
413 * CONNECT requests. The request line was missing.
414 * - Add two more XXX reminders as we don't have enough already.
416 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
417 * As Hal noticed, checking dispatch_cgi() as the last cruncher
418 * looks like a bug if CGI requests are blocked unintentionally,
419 * so don't do it unless the user enabled the new config option
420 * "allow-cgi-request-crunching".
422 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
423 * - Allow to rewrite the request destination behind the client's back.
424 * - Turn the weird-looking unconditional for loop that
425 * reads the client request into a conditional while loop.
426 * Move the stuff that only runs once out of the loop.
427 * - Move parts of chat(), server_content_type() and the
428 * necessary stuff to fix BR#1750917 into get_filter_function().
430 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
431 * Temporary workaround for BR#1730105.
433 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
434 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
435 * deadlocks and crashes on OpenBSD and possibly other OS with neither
436 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
437 * Thanks to Ralf Horstmann for report and solution.
439 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
440 * Add forward-override{} to change the forwarding settings through
441 * action sections. This is mainly interesting to forward different
442 * clients differently (for example based on User-Agent or request
445 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
446 * - Let usage() mention the --chroot parameter.
447 * - Use read_socket() consistently and always leave
448 * the last buffer byte alone, even in cases where
449 * null termination (currently) doesn't matter.
451 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
452 * - Fix config file loading on Unix if no config file is specified.
453 * Since r1.97 Privoxy would always interpret the last argument as
454 * config file, even if it's a valid command line option.
455 * - Abort in case of unrecognized command line options. Closes #1719696.
456 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
457 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
459 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
460 * - Don't rerun crunchers that only depend on the request URL.
461 * - Don't count redirects and CGI requests as "blocked requests".
463 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
464 * Support crunching based on tags created by server-header taggers.
466 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
467 * Make HTTP snippets static (again). Add a Content-Type for those
468 * with content so the browser doesn't guess it based on the URL.
470 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
471 * Move crunching and request line rebuilding out of chat().
473 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
474 * Introduce tags as alternative way to specify which
475 * actions apply to a request. At the moment tags can be
476 * created based on client and server headers.
478 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
479 * Don't CLF-log CONNECT requests twice.
481 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
482 * Log the source address for ACL-related connection drops.
484 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
485 * New config option: enforce-blocks.
487 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
488 * - Move null byte check into separate function.
489 * - Don't confuse the client with error pages
490 * if a CONNECT request was already confirmed.
492 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
493 * Speed up NULL byte escaping and only log the complete
494 * NULL byte requests with header debugging enabled.
496 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
497 * Answer requests that contain NULL bytes with
498 * a custom response instead of waiting for more
499 * data until the client eventually hangs up.
501 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
502 * - Move delivery and logging of crunched responses
503 * from chat() into send_crunch_response().
504 * - Display the reason for generating http_responses.
505 * - Log the content length for LOG_LEVEL_CLF correctly
506 * (still incorrect for some fixed responses).
507 * - Reword an incorrect comment about
508 * treat-forbidden-connects-like-blocks violating
510 * - Add some log messages.
512 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
513 * Move mutex initialization into separate
514 * function and exit in case of errors.
516 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
517 * - Start to reduce chat()'s line count and move
518 * parts of it into separate functions.
519 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
521 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
522 * - Add Proxy-Agent header to HTTP snippets that are
523 * supposed to reach HTTP clients only.
524 * - Made a few CONNECT log messages more descriptive.
525 * - Catch completely empty server responses (as seen
526 * with Tor's fake ".noconnect" top level domain).
527 * - Use shiny new "forwarding-failed" template for socks errors.
529 * Revision 1.118 2007/01/07 07:43:43 joergs
530 * AmigaOS4 support added.
532 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
533 * Added config option accept-intercepted-requests
534 * and disabled it by default.
536 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
537 * Reverted parts of my last commit
538 * to keep error handling working.
540 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
541 * Fixed gcc43 conversion warnings.
543 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
544 * Fix -pedantic ISO C warning about converting
545 * from function pointer to object pointer.
547 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
548 * Silence compiler warning I introduced with my last commit.
550 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
551 * Mutex protect rand() if POSIX threading
552 * is used, warn the user if that's not possible
553 * and stop using it on _WIN32 where it could
556 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
557 * Don't prevent core dumps by catching SIGABRT.
558 * It's rude and makes debugging unreasonable painful.
560 * Revision 1.110 2006/12/13 14:52:53 etresoft
561 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
563 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
564 * Privoxy is now able to run as intercepting
565 * proxy in combination with any packet filter
566 * that does the port redirection. The destination
567 * is extracted from the "Host:" header which
568 * should be available for nearly all requests.
570 * Moved HTTP snipplets into jcc.c.
571 * Added error message for gopher proxy requests.
573 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
574 * Only unlink the pidfile if it's actually used.
576 * Change order of interception checks to make
577 * it possible to block or redirect requests for
580 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
581 * Make pthread mutex locking more generic. Instead of
582 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
583 * and use mutex locking unless there is an _r function
584 * available. Better safe than sorry.
586 * Fixes "./configure --disable-pthread" and should result
587 * in less threading-related problems on pthread-using platforms,
588 * but it still doesn't fix BR#1122404.
590 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
591 * Move pthread.h inclusion from jcc.c to jcc.h.
592 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
594 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
595 * Don't exit after receiving the second SIGHUP on Solaris.
597 * Fixes BR 1052235, but the same problem may exist on other
598 * systems. Once 3.0.6 is out we should use sigset()
599 * where available and see if it breaks anything.
601 * Revision 1.104 2006/09/23 13:26:38 roro
602 * Replace TABs by spaces in source code.
604 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
605 * Fix +redirect{}. Didn't work with -fast-redirects.
607 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
608 * Respond with 400 and a short text message
609 * if the client tries to use Privoxy as FTP proxy.
611 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
612 * Make number of retries in case of forwarded-connect problems
613 * a config file option (forwarded-connect-retries) and use 0 as
616 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
617 * Set random(3) seed.
619 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
620 * Follow the OpenBSD port's lead and protect the resolve
621 * functions on OpenBSD as well.
623 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
624 * --user fix. Only use the user as group if no group is specified.
625 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
627 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
628 * Windows service (re-)integration
632 * --install[:service_name]
633 * --uninstall[:service_name]
636 * They work as follows:
637 * --install will create a service for you and then terminate.
638 * By default the service name will be "privoxy" (without the quotes).
639 * However you can run multiple services if you wish, just by adding
640 * a colon and then a name (no spaces).
642 * --uninstall follows the exact same rules a --install.
644 * --service is used when the program is executed by the service
645 * control manager, and in normal circumstances would never be
646 * used as a command line argument.
648 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
649 * Windows service integration
651 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
652 * Incorporate Fabian Keil's patch work:
653 http://www.fabiankeil.de/sourcecode/privoxy/
655 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
656 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
657 * with what was really the latest development (the v_3_0_branch branch)
659 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
660 * Thanks to Jindrich Makovicka for a race condition fix for the log
661 * file. The race condition remains for non-pthread implementations.
662 * Reference patch #1175720.
664 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
665 * Add the ability to check jpeg images for invalid
666 * lengths of comment blocks. Defensive strategy
667 * against the exploit:
668 * Microsoft Security Bulletin MS04-028
669 * Buffer Overrun in JPEG Processing (GDI+) Could
670 * Allow Code Execution (833987)
671 * Enabled with +inspect-jpegs in actions files.
673 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
674 * - Fixed usage info for non-unix platforms
675 * - Fixed small cmdline parsing bug
677 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
678 * Diagnostics: Now preserve the returncode of pthread_create
679 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
681 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
682 * No longer ignore SIGCHLD. Fixes bug #769381
684 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
685 * Close jarfile on graceful exit, remove stray line
687 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
688 * Cosmetics: Killed a warning, a typo and an allocation left at exit
690 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
691 * No longer rely on non-POSIX.1 extensions of getcwd().
694 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
695 * Replaced setenv() by posix-compliant putenv()
696 * Thanks to Neil McCalden (nmcc AT users.sf.net).
698 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
699 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
701 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
702 * Clean-up and extension of improvements for forked mode:
703 * - Child's return code now consists of flags RC_FLAG_*
704 * - Reporting toggle to parent now properly #ifdef'ed
705 * - Children now report blocking to parent. This enables
706 * statistics in forked mode
708 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
709 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
710 * Children now report having been toggled through _exit(17), parents
711 * watch for that code and toggle themselves if found.
713 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
714 * Wrapping all *_r functions (the non-_r versions of them) with
715 * mutex semaphores for OSX. Hopefully this will take care of all
716 * of those pesky crash reports.
718 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
719 * Fixed two mostly harmless mem leaks
721 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
722 * Fix: Head of global clients list now initialized to NULL
724 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
725 * Added basic support for OPTIONS and TRACE HTTP methods:
726 * - New interceptor direct_response() added in chat().
727 * - sed() moved to earlier in the process, so that the
728 * Host: header is evaluated before actions and forwarding
731 * Revision 1.92 2002/05/08 16:00:46 oes
732 * Chat's buffer handling:
733 * - Fixed bug with unchecked out-of-mem conditions
734 * while reading client request & server headers
735 * - No longer predict if the buffer limit will be exceeded
736 * in the next read -- check add_to_iob's new
737 * return code. If buffer couldn't be extended
738 * (policy or out-of-mem) while
739 * - reading from client: abort
740 * - reading server headers: send error page
741 * - buffering server body for filter: flush,
742 * and if that fails: send error page
744 * Revision 1.91 2002/04/08 20:35:58 swa
747 * Revision 1.90 2002/04/02 14:57:28 oes
748 * Made sending wafers independent of FEATURE_COOKIE_JAR
750 * Revision 1.89 2002/03/31 17:18:59 jongfoster
751 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
753 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
754 * More compiler warning message maintenance
756 * Revision 1.87 2002/03/26 22:29:54 swa
757 * we have a new homepage!
759 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
760 * Workaround for closing the jarfile before load_config() comes around again
762 * Revision 1.85 2002/03/24 15:23:33 jongfoster
765 * Revision 1.84 2002/03/24 13:25:43 swa
766 * name change related issues
768 * Revision 1.83 2002/03/16 23:54:06 jongfoster
769 * Adding graceful termination feature, to help look for memory leaks.
770 * If you enable this (which, by design, has to be done by hand
771 * editing config.h) and then go to http://i.j.b/die, then the program
772 * will exit cleanly after the *next* request. It should free all the
773 * memory that was used.
775 * Revision 1.82 2002/03/13 00:27:05 jongfoster
778 * Revision 1.81 2002/03/12 01:42:50 oes
779 * Introduced modular filters
781 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
782 * OS/2 port maintenance:
783 * - Fixed EMX build - it had decayed a little
784 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
785 * substituted a memset for now.
787 * Revision 1.79 2002/03/09 20:03:52 jongfoster
788 * - Making various functions return int rather than size_t.
789 * (Undoing a recent change). Since size_t is unsigned on
790 * Windows, functions like read_socket that return -1 on
791 * error cannot return a size_t.
793 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
794 * crashes, and also frequently caused JB to jump to 100%
795 * CPU and stay there. (Because it thought it had just
796 * read ((unsigned)-1) == 4Gb of data...)
798 * - The signature of write_socket has changed, it now simply
799 * returns success=0/failure=nonzero.
801 * - Trying to get rid of a few warnings --with-debug on
802 * Windows, I've introduced a new type "jb_socket". This is
803 * used for the socket file descriptors. On Windows, this
804 * is SOCKET (a typedef for unsigned). Everywhere else, it's
805 * an int. The error value can't be -1 any more, so it's
806 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
807 * Windows it maps to the #define INVALID_SOCKET.)
809 * - The signature of bind_port has changed.
811 * Revision 1.78 2002/03/08 21:35:04 oes
812 * Added optional group supplement to --user option. Will now use default group of user if no group given
814 * Revision 1.77 2002/03/07 03:52:06 oes
815 * - Fixed compiler warnings etc
816 * - Improved handling of failed DNS lookups
818 * Revision 1.76 2002/03/06 22:54:35 jongfoster
819 * Automated function-comment nitpicking.
821 * Revision 1.75 2002/03/06 10:02:19 oes
822 * Fixed stupid bug when --user was not given
824 * Revision 1.74 2002/03/06 00:49:31 jongfoster
825 * Fixing warning on Windows
826 * Making #ifdefs that refer to the same variable consistently
827 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
829 * Revision 1.73 2002/03/05 23:57:30 hal9
830 * Stray character 's' on line 1618 was breaking build.
832 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
833 * - Re-enable OS/2 building after new parms were added
834 * - Fix false out of memory report when resolving CGI templates when no IP
835 * address is available of failed attempt (a la no such domain)
837 * Revision 1.71 2002/03/05 18:13:56 oes
838 * Added --user option
840 * Revision 1.70 2002/03/05 04:52:42 oes
841 * Deleted non-errlog debugging code
843 * Revision 1.69 2002/03/04 23:50:00 jongfoster
844 * Splitting off bind_port() call into bind_port_helper(), with
847 * Revision 1.68 2002/03/04 20:17:32 oes
850 * Revision 1.67 2002/03/04 18:18:57 oes
851 * - Removed _DEBUG mode
852 * - Cleand up cmdline parsing
853 * - Introduced --no-daemon, --pidfile options
854 * - Cleaned up signal handling:
855 * - Terminate cleanly on INT, TERM and ABRT
856 * - Schedule logfile for re-opening on HUP
857 * - Ignore CHLD and PIPE
858 * - Leave the rest with their default handlers
859 * - Uniform handler registration
860 * - Added usage() function
861 * - Played styleguide police
863 * Revision 1.66 2002/03/03 15:06:55 oes
864 * Re-enabled automatic config reloading
866 * Revision 1.65 2002/03/03 14:49:11 oes
867 * Fixed CLF logging: Now uses client's original HTTP request
869 * Revision 1.64 2002/03/03 09:18:03 joergs
870 * Made jumbjuster work on AmigaOS again.
872 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
873 * Clean up a little CRLF unpleasantness that suddenly appeared
875 * Revision 1.62 2002/02/20 23:17:23 jongfoster
876 * Detecting some out-of memory conditions and exiting with a log message.
878 * Revision 1.61 2002/01/17 21:01:52 jongfoster
879 * Moving all our URL and URL pattern parsing code to urlmatch.c.
881 * Revision 1.60 2001/12/30 14:07:32 steudten
882 * - Add signal handling (unix)
883 * - Add SIGHUP handler (unix)
884 * - Add creation of pidfile (unix)
885 * - Add action 'top' in rc file (RH)
886 * - Add entry 'SIGNALS' to manpage
887 * - Add exit message to logfile (unix)
889 * Revision 1.59 2001/12/13 14:07:18 oes
890 * Fixed Bug: 503 error page now sent OK
892 * Revision 1.58 2001/11/30 23:37:24 jongfoster
893 * Renaming the Win32 config file to config.txt - this is almost the
894 * same as the corresponding UNIX name "config"
896 * Revision 1.57 2001/11/16 00:47:43 jongfoster
897 * Changing the tty-disconnection code to use setsid().
899 * Revision 1.56 2001/11/13 20:20:54 jongfoster
900 * Tabs->spaces, fixing a bug with missing {} around an if()
902 * Revision 1.55 2001/11/13 20:14:53 jongfoster
903 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
905 * Revision 1.54 2001/11/07 00:03:14 steudten
906 * Give reliable return value if an error
907 * occurs not just 0 with new daemon mode.
909 * Revision 1.53 2001/11/05 21:41:43 steudten
910 * Add changes to be a real daemon just for unix os.
911 * (change cwd to /, detach from controlling tty, set
912 * process group and session leader to the own process.
914 * Add some fatal-error log message for failed malloc().
915 * Add '-d' if compiled with 'configure --with-debug' to
916 * enable debug output.
918 * Revision 1.52 2001/10/26 20:11:20 jongfoster
919 * Fixing type mismatch
921 * Revision 1.51 2001/10/26 17:38:28 oes
924 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
925 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
926 * threads to call select() simultaneously. So, it's time to do a real, live,
927 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
928 * (native). Both versions will work, but using __OS2__ offers multi-threading.
930 * Revision 1.49 2001/10/23 21:41:35 jongfoster
931 * Added call to initialize the (statically-allocated of course)
932 * "out of memory" CGI response.
934 * Revision 1.48 2001/10/10 19:56:46 jongfoster
935 * Moving some code that wasn't cookie-related out of an #ifdef
938 * Revision 1.47 2001/10/10 16:44:36 oes
939 * Added CONNECT destination port limitation check
941 * Revision 1.46 2001/10/08 15:17:41 oes
942 * Re-enabled SSL forwarding
944 * Revision 1.45 2001/10/07 15:42:11 oes
945 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
947 * Moved downgrading of the HTTP version from parse_http_request to
948 * chat(), since we can't decide if it is necessary before we have
949 * determined the actions for the URL. The HTTP command is now
950 * *always* re-built so the repairs need no longer be special-cased.
952 * filter_popups now gets a csp pointer so it can raise the new
953 * CSP_FLAG_MODIFIED flag.
957 * Added configurable size limit for the IOB. If the IOB grows so
958 * large that the next read would exceed the limit, the header
959 * is generated, and the header & unfiltered buffer are flushed
960 * to the client. Chat then continues in non-buffering,
961 * non-filtering body mode.
963 * Revision 1.44 2001/10/02 18:13:57 oes
966 * Revision 1.43 2001/10/02 15:32:13 oes
967 * Moved generation of hdr
969 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
970 * Cleaning up 2 compiler warnings on OS/2.
972 * Revision 1.41 2001/09/16 17:05:14 jongfoster
973 * Removing unused #include showarg.h
975 * Revision 1.40 2001/09/16 15:41:45 jongfoster
976 * Fixing signed/unsigned comparison warning.
978 * Revision 1.39 2001/09/16 13:21:27 jongfoster
979 * Changes to use new list functions.
981 * Revision 1.38 2001/09/16 13:01:46 jongfoster
982 * Removing redundant function call that zeroed zalloc()'d memory.
984 * Revision 1.37 2001/09/10 11:12:24 oes
985 * Deleted unused variable
987 * Revision 1.36 2001/09/10 10:56:15 oes
988 * Silenced compiler warnings
990 * Revision 1.35 2001/07/31 14:44:22 oes
991 * Deleted unused size parameter from filter_popups()
993 * Revision 1.34 2001/07/30 22:08:36 jongfoster
994 * Tidying up #defines:
995 * - All feature #defines are now of the form FEATURE_xxx
996 * - Permanently turned off WIN_GUI_EDIT
997 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
999 * Revision 1.33 2001/07/29 19:32:00 jongfoster
1000 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
1002 * Revision 1.32 2001/07/29 18:47:05 jongfoster
1003 * Adding missing #include "loadcfg.h"
1005 * Revision 1.31 2001/07/29 12:17:48 oes
1006 * Applied pthread fix by Paul Lieverse
1008 * Revision 1.30 2001/07/25 22:57:13 jongfoster
1009 * __BEOS__ no longer overrides FEATURE_PTHREAD.
1010 * This is because FEATURE_PTHREAD will soon be widely used, so I
1011 * want to keep it simple.
1013 * Revision 1.29 2001/07/24 12:47:06 oes
1014 * Applied BeOS support update by Eugenia
1016 * Revision 1.28 2001/07/23 13:26:12 oes
1017 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
1019 * Revision 1.27 2001/07/19 19:09:47 haroon
1020 * - Added code to take care of the situation where while processing the first
1021 * server response (which includes the server header), after finding the end
1022 * of the headers we were not looking past the end of the headers for
1023 * content modification. I enabled it for filter_popups.
1024 * Someone else should look to see if other similar operations should be
1025 * done to the discarded portion of the buffer.
1027 * Note 2001/07/20: No, the other content modification mechanisms will process
1028 * the whole iob later anyway. --oes
1030 * Revision 1.26 2001/07/18 12:31:36 oes
1033 * Revision 1.25 2001/07/15 19:43:49 jongfoster
1034 * Supports POSIX threads.
1035 * Also removed some unused #includes.
1037 * Revision 1.24 2001/07/13 14:00:40 oes
1038 * - Generic content modification scheme:
1039 * Each feature has its own applicability flag that is set
1040 * from csp->action->flags.
1041 * Replaced the "filtering" int flag , by a function pointer
1042 * "content_filter" to the function that will do the content
1043 * modification. If it is != NULL, the document will be buffered
1044 * and processed through *content_filter, which must set
1045 * csp->content_length and return a modified copy of the body
1046 * or return NULL (on failiure).
1047 * - Changed csp->is_text to the more generic bitmap csp->content_type
1048 * which can currently take the valued CT_TEXT or CT_GIF
1049 * - Reformatting etc
1050 * - Removed all #ifdef PCRS
1052 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
1053 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
1054 * the `block_acl' call. This prevents a compilation error when the user
1055 * does not wish to use the "ACL" feature.
1057 * Revision 1.22 2001/06/29 21:45:41 oes
1058 * Indentation, CRLF->LF, Tab-> Space
1060 * Revision 1.21 2001/06/29 13:29:36 oes
1061 * - Cleaned up, improved comments
1062 * - Unified all possible interceptors (CGI,
1063 * block, trust, fast_redirect) in one
1064 * place, with one (CGI) answer generation
1065 * mechansim. Much clearer now.
1066 * - Removed the GIF image generation, which
1067 * is now done in filters.c:block_url()
1068 * - Made error conditions like domain lookup
1069 * failiure or (various) problems while talking
1070 * to the server use cgi.c:error_response()
1071 * instead of generating HTML/HTTP in chat() (yuck!)
1072 * - Removed logentry from cancelled commit
1074 * Revision 1.20 2001/06/09 10:55:28 jongfoster
1075 * Changing BUFSIZ ==> BUFFER_SIZE
1077 * Revision 1.19 2001/06/07 23:12:52 jongfoster
1078 * Replacing function pointer in struct gateway with a directly
1079 * called function forwarded_connect().
1080 * Replacing struct gateway with struct forward_spec
1082 * Revision 1.18 2001/06/03 19:12:16 oes
1083 * introduced new cgi handling
1085 * Revision 1.17 2001/06/01 20:07:23 jongfoster
1086 * Now uses action +image-blocker{} rather than config->tinygif
1088 * Revision 1.16 2001/06/01 18:49:17 jongfoster
1089 * Replaced "list_share" with "list" - the tiny memory gain was not
1090 * worth the extra complexity.
1092 * Revision 1.15 2001/05/31 21:24:47 jongfoster
1093 * Changed "permission" to "action" throughout.
1094 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
1095 * Moved vanilla wafer check into chat(), since we must now
1096 * decide whether or not to add it based on the URL.
1098 * Revision 1.14 2001/05/29 20:14:01 joergs
1099 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
1102 * Revision 1.13 2001/05/29 09:50:24 jongfoster
1103 * Unified blocklist/imagelist/permissionslist.
1104 * File format is still under discussion, but the internal changes
1105 * are (mostly) done.
1107 * Also modified interceptor behaviour:
1108 * - We now intercept all URLs beginning with one of the following
1109 * prefixes (and *only* these prefixes):
1111 * * http://ijbswa.sf.net/config/
1112 * * http://ijbswa.sourceforge.net/config/
1113 * - New interceptors "home page" - go to http://i.j.b/ to see it.
1114 * - Internal changes so that intercepted and fast redirect pages
1115 * are not replaced with an image.
1116 * - Interceptors now have the option to send a binary page direct
1117 * to the client. (i.e. ijb-send-banner uses this)
1118 * - Implemented show-url-info interceptor. (Which is why I needed
1119 * the above interceptors changes - a typical URL is
1120 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
1121 * The previous mechanism would not have intercepted that, and
1122 * if it had been intercepted then it then it would have replaced
1123 * it with an image.)
1125 * Revision 1.12 2001/05/27 22:17:04 oes
1127 * - re_process_buffer no longer writes the modified buffer
1128 * to the client, which was very ugly. It now returns the
1129 * buffer, which it is then written by chat.
1131 * - content_length now adjusts the Content-Length: header
1132 * for modified documents rather than crunch()ing it.
1133 * (Length info in csp->content_length, which is 0 for
1134 * unmodified documents)
1136 * - For this to work, sed() is called twice when filtering.
1138 * Revision 1.11 2001/05/26 17:27:53 jongfoster
1139 * Added support for CLF and fixed LOG_LEVEL_LOG.
1140 * Also did CRLF->LF fix of my previous patch.
1142 * Revision 1.10 2001/05/26 15:26:15 jongfoster
1143 * ACL feature now provides more security by immediately dropping
1144 * connections from untrusted hosts.
1146 * Revision 1.9 2001/05/26 00:28:36 jongfoster
1147 * Automatic reloading of config file.
1148 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
1149 * Most of the global variables have been moved to a new
1150 * struct configuration_spec, accessed through csp->config->globalname
1151 * Most of the globals remaining are used by the Win32 GUI.
1153 * Revision 1.8 2001/05/25 22:43:18 jongfoster
1154 * Fixing minor memory leak and buffer overflow.
1156 * Revision 1.7 2001/05/25 22:34:30 jongfoster
1159 * Revision 1.6 2001/05/23 00:13:58 joergs
1160 * AmigaOS support fixed.
1162 * Revision 1.5 2001/05/22 18:46:04 oes
1164 * - Enabled filtering banners by size rather than URL
1165 * by adding patterns that replace all standard banner
1166 * sizes with the "Junkbuster" gif to the re_filterfile
1168 * - Enabled filtering WebBugs by providing a pattern
1169 * which kills all 1x1 images
1171 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
1172 * which is selected by the (nonstandard and therefore
1173 * capital) letter 'U' in the option string.
1174 * It causes the quantifiers to be ungreedy by default.
1175 * Appending a ? turns back to greedy (!).
1177 * - Added a new interceptor ijb-send-banner, which
1178 * sends back the "Junkbuster" gif. Without imagelist or
1179 * MSIE detection support, or if tinygif = 1, or the
1180 * URL isn't recognized as an imageurl, a lame HTML
1181 * explanation is sent instead.
1183 * - Added new feature, which permits blocking remote
1184 * script redirects and firing back a local redirect
1186 * The feature is conditionally compiled, i.e. it
1187 * can be disabled with --disable-fast-redirects,
1188 * plus it must be activated by a "fast-redirects"
1189 * line in the config file, has its own log level
1190 * and of course wants to be displayed by show-proxy-args
1191 * Note: Boy, all the #ifdefs in 1001 locations and
1192 * all the fumbling with configure.in and acconfig.h
1193 * were *way* more work than the feature itself :-(
1195 * - Because a generic redirect template was needed for
1196 * this, tinygif = 3 now uses the same.
1198 * - Moved GIFs, and other static HTTP response templates
1201 * - Some minor fixes
1203 * - Removed some >400 CRs again (Jon, you really worked
1206 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1207 * Made failure to bind() a fatal error.
1209 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1210 * Version 2.9.4 checkin.
1211 * - Merged popupfile and cookiefile, and added control over PCRS
1212 * filtering, in new "permissionsfile".
1213 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1214 * file error you now get a message box (in the Win32 GUI) rather
1215 * than the program exiting with no explanation.
1216 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1218 * - Removed tabs from "config"
1219 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1220 * - Bumped up version number.
1222 * Revision 1.2 2001/05/17 22:34:44 oes
1223 * - Added hint on GIF char array generation to jcc.c
1224 * - Cleaned CRLF's from the sources and related files
1225 * - Repaired logging for REF and FRC
1227 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1228 * Initial import of version 2.9.3 source tree
1231 *********************************************************************/
1237 #include <sys/types.h>
1246 # ifndef FEATURE_PTHREAD
1250 # include <windows.h>
1251 # include <process.h>
1252 # endif /* ndef FEATURE_PTHREAD */
1255 # ifndef _WIN_CONSOLE
1256 # include "w32log.h"
1257 # endif /* ndef _WIN_CONSOLE */
1258 # include "w32svrapi.h"
1260 #else /* ifndef _WIN32 */
1262 # if !defined (__OS2__)
1263 # include <unistd.h>
1264 # include <sys/wait.h>
1265 # endif /* ndef __OS2__ */
1266 # include <sys/time.h>
1267 # include <sys/stat.h>
1268 # include <sys/ioctl.h>
1271 #include <sys/termios.h>
1279 # include <signal.h>
1282 # include <socket.h> /* BeOS has select() for sockets only. */
1283 # include <OS.h> /* declarations for threads and stuff. */
1286 # if defined(__EMX__) || defined(__OS2__)
1287 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1292 #define bzero(B,N) memset(B,0x00,n)
1296 # include <select.h>
1301 #include "project.h"
1304 #include "filters.h"
1305 #include "loaders.h"
1306 #include "parsers.h"
1307 #include "miscutil.h"
1309 #include "jbsockets.h"
1310 #include "gateway.h"
1311 #include "actions.h"
1313 #include "loadcfg.h"
1314 #include "urlmatch.h"
1316 const char jcc_h_rcs[] = JCC_H_VERSION;
1317 const char project_h_rcs[] = PROJECT_H_VERSION;
1320 struct client_state clients[1];
1321 struct file_list files[1];
1323 #ifdef FEATURE_STATISTICS
1324 int urls_read = 0; /* total nr of urls read inc rejected */
1325 int urls_rejected = 0; /* total nr of urls rejected */
1326 #endif /* def FEATURE_STATISTICS */
1328 #ifdef FEATURE_GRACEFUL_TERMINATION
1329 int g_terminate = 0;
1332 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1333 static void sig_handler(int the_signal);
1335 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1336 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1337 static jb_err get_server_headers(struct client_state *csp);
1338 static const char *crunch_reason(const struct http_response *rsp);
1339 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1340 static char *get_request_line(struct client_state *csp);
1341 static jb_err receive_client_request(struct client_state *csp);
1342 static jb_err parse_client_request(struct client_state *csp);
1343 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1344 static jb_err change_request_destination(struct client_state *csp);
1345 static void chat(struct client_state *csp);
1346 static void serve(struct client_state *csp);
1347 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1348 static void usage(const char *myname);
1350 static void initialize_mutexes(void);
1351 static jb_socket bind_port_helper(struct configuration_spec *config);
1352 static void listen_loop(void);
1355 void serve(struct client_state *csp);
1356 #else /* ifndef AMIGA */
1357 static void serve(struct client_state *csp);
1358 #endif /* def AMIGA */
1361 static int32 server_thread(void *data);
1362 #endif /* def __BEOS__ */
1365 #define sleep(N) Sleep(((N) * 1000))
1369 #define sleep(N) DosSleep(((N) * 100))
1372 #ifdef MUTEX_LOCKS_AVAILABLE
1374 * XXX: Does the locking stuff really belong in this file?
1376 privoxy_mutex_t log_mutex;
1377 privoxy_mutex_t log_init_mutex;
1378 privoxy_mutex_t connection_reuse_mutex;
1380 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1381 privoxy_mutex_t resolver_mutex;
1382 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1384 #ifndef HAVE_GMTIME_R
1385 privoxy_mutex_t gmtime_mutex;
1386 #endif /* ndef HAVE_GMTIME_R */
1388 #ifndef HAVE_LOCALTIME_R
1389 privoxy_mutex_t localtime_mutex;
1390 #endif /* ndef HAVE_GMTIME_R */
1393 privoxy_mutex_t rand_mutex;
1394 #endif /* ndef HAVE_RANDOM */
1396 #endif /* def MUTEX_LOCKS_AVAILABLE */
1399 const char *basedir = NULL;
1400 const char *pidfile = NULL;
1401 static int received_hup_signal = 0;
1402 #endif /* defined unix */
1404 /* HTTP snipplets. */
1405 static const char CSUCCEED[] =
1406 "HTTP/1.0 200 Connection established\r\n"
1407 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1409 static const char CHEADER[] =
1410 "HTTP/1.0 400 Invalid header received from client\r\n"
1411 "Proxy-Agent: Privoxy " VERSION "\r\n"
1412 "Content-Type: text/plain\r\n"
1413 "Connection: close\r\n\r\n"
1414 "Invalid header received from client.\r\n";
1416 static const char FTP_RESPONSE[] =
1417 "HTTP/1.0 400 Invalid request received from client\r\n"
1418 "Content-Type: text/plain\r\n"
1419 "Connection: close\r\n\r\n"
1420 "Invalid request. Privoxy doesn't support FTP.\r\n";
1422 static const char GOPHER_RESPONSE[] =
1423 "HTTP/1.0 400 Invalid request received from client\r\n"
1424 "Content-Type: text/plain\r\n"
1425 "Connection: close\r\n\r\n"
1426 "Invalid request. Privoxy doesn't support gopher.\r\n";
1428 /* XXX: should be a template */
1429 static const char MISSING_DESTINATION_RESPONSE[] =
1430 "HTTP/1.0 400 Bad request received from client\r\n"
1431 "Proxy-Agent: Privoxy " VERSION "\r\n"
1432 "Content-Type: text/plain\r\n"
1433 "Connection: close\r\n\r\n"
1434 "Bad request. Privoxy was unable to extract the destination.\r\n";
1436 /* XXX: should be a template */
1437 static const char NO_SERVER_DATA_RESPONSE[] =
1438 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1439 "Proxy-Agent: Privoxy " VERSION "\r\n"
1440 "Content-Type: text/plain\r\n"
1441 "Connection: close\r\n\r\n"
1442 "Empty server or forwarder response.\r\n"
1443 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1445 /* XXX: should be a template */
1446 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1447 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1448 "Proxy-Agent: Privoxy " VERSION "\r\n"
1449 "Content-Type: text/plain\r\n"
1450 "Connection: close\r\n\r\n"
1451 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1454 /* XXX: should be a template */
1455 static const char NULL_BYTE_RESPONSE[] =
1456 "HTTP/1.0 400 Bad request received from client\r\n"
1457 "Proxy-Agent: Privoxy " VERSION "\r\n"
1458 "Content-Type: text/plain\r\n"
1459 "Connection: close\r\n\r\n"
1460 "Bad request. Null byte(s) before end of request.\r\n";
1463 /* XXX: should be a template */
1464 static const char MESSED_UP_REQUEST_RESPONSE[] =
1465 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1466 "Proxy-Agent: Privoxy " VERSION "\r\n"
1467 "Content-Type: text/plain\r\n"
1468 "Connection: close\r\n\r\n"
1469 "Bad request. Messed up with header filters.\r\n";
1471 /* XXX: should be a template */
1472 static const char CONNECTION_TIMEOUT_RESPONSE[] =
1473 "HTTP/1.0 502 Connection timeout\r\n"
1474 "Proxy-Agent: Privoxy " VERSION "\r\n"
1475 "Content-Type: text/plain\r\n"
1476 "Connection: close\r\n\r\n"
1477 "The connection timed out.\r\n";
1479 /* A function to crunch a response */
1480 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1482 /* Crunch function flags */
1483 #define CF_NO_FLAGS 0
1484 /* Cruncher applies to forced requests as well */
1485 #define CF_IGNORE_FORCE 1
1486 /* Crunched requests are counted for the block statistics */
1487 #define CF_COUNT_AS_REJECT 2
1489 /* A crunch function and its flags */
1492 const crunch_func_ptr cruncher;
1496 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1498 /* Complete list of cruncher functions */
1499 static const struct cruncher crunchers_all[] = {
1500 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1501 { block_url, CF_COUNT_AS_REJECT },
1502 #ifdef FEATURE_TRUST
1503 { trust_url, CF_COUNT_AS_REJECT },
1504 #endif /* def FEATURE_TRUST */
1505 { redirect_url, CF_NO_FLAGS },
1506 { dispatch_cgi, CF_IGNORE_FORCE},
1510 /* Light version, used after tags are applied */
1511 static const struct cruncher crunchers_light[] = {
1512 { block_url, CF_COUNT_AS_REJECT },
1513 { redirect_url, CF_NO_FLAGS },
1519 * XXX: Don't we really mean
1525 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1526 /*********************************************************************
1528 * Function : sig_handler
1530 * Description : Signal handler for different signals.
1531 * Exit gracefully on TERM and INT
1532 * or set a flag that will cause the errlog
1533 * to be reopened by the main thread on HUP.
1536 * 1 : the_signal = the signal cause this function to call
1540 *********************************************************************/
1541 static void sig_handler(int the_signal)
1547 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1559 received_hup_signal = 1;
1565 * We shouldn't be here, unless we catch signals
1566 * in main() that we can't handle here!
1568 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1576 /*********************************************************************
1578 * Function : client_protocol_is_unsupported
1580 * Description : Checks if the client used a known unsupported
1581 * protocol and deals with it by sending an error
1585 * 1 : csp = Current client state (buffers, headers, etc...)
1586 * 2 : req = the first request line send by the client
1588 * Returns : TRUE if an error response has been generated, or
1589 * FALSE if the request doesn't look invalid.
1591 *********************************************************************/
1592 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1595 * If it's a FTP or gopher request, we don't support it.
1597 * These checks are better than nothing, but they might
1598 * not work in all configurations and some clients might
1599 * have problems digesting the answer.
1601 * They should, however, never cause more problems than
1602 * Privoxy's old behaviour (returning the misleading HTML
1605 * "Could not resolve http://(ftp|gopher)://example.org").
1607 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1609 const char *response = NULL;
1610 const char *protocol = NULL;
1612 if (!strncmpic(req, "GET ftp://", 10))
1614 response = FTP_RESPONSE;
1619 response = GOPHER_RESPONSE;
1620 protocol = "GOPHER";
1622 log_error(LOG_LEVEL_ERROR,
1623 "%s tried to use Privoxy as %s proxy: %s",
1624 csp->ip_addr_str, protocol, req);
1625 log_error(LOG_LEVEL_CLF,
1626 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1628 write_socket(csp->cfd, response, strlen(response));
1637 /*********************************************************************
1639 * Function : get_request_destination_elsewhere
1641 * Description : If the client's request was redirected into
1642 * Privoxy without the client's knowledge,
1643 * the request line lacks the destination host.
1645 * This function tries to get it elsewhere,
1646 * provided accept-intercepted-requests is enabled.
1648 * "Elsewhere" currently only means "Host: header",
1649 * but in the future we may ask the redirecting
1650 * packet filter to look the destination up.
1652 * If the destination stays unknown, an error
1653 * response is send to the client and headers
1654 * are freed so that chat() can return directly.
1657 * 1 : csp = Current client state (buffers, headers, etc...)
1658 * 2 : headers = a header list
1660 * Returns : JB_ERR_OK if the destination is now known, or
1661 * JB_ERR_PARSE if it isn't.
1663 *********************************************************************/
1664 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1668 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1670 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1671 " Privoxy isn't configured to accept intercepted requests.",
1672 csp->ip_addr_str, csp->http->cmd);
1673 /* XXX: Use correct size */
1674 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1675 csp->ip_addr_str, csp->http->cmd);
1677 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1678 destroy_list(headers);
1680 return JB_ERR_PARSE;
1682 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1684 /* Split the domain we just got for pattern matching */
1685 init_domain_components(csp->http);
1691 /* We can't work without destination. Go spread the news.*/
1693 req = list_to_text(headers);
1695 /* XXX: Use correct size */
1696 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1697 csp->ip_addr_str, csp->http->cmd);
1698 log_error(LOG_LEVEL_ERROR,
1699 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1700 csp->ip_addr_str, csp->http->cmd, req);
1703 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1704 destroy_list(headers);
1706 return JB_ERR_PARSE;
1709 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1710 * to get the destination IP address, use it as host directly
1711 * or do a reverse DNS lookup first.
1716 /*********************************************************************
1718 * Function : get_server_headers
1720 * Description : Parses server headers in iob and fills them
1721 * into csp->headers so that they can later be
1725 * 1 : csp = Current client state (buffers, headers, etc...)
1727 * Returns : JB_ERR_OK if everything went fine, or
1728 * JB_ERR_PARSE if the headers were incomplete.
1730 *********************************************************************/
1731 static jb_err get_server_headers(struct client_state *csp)
1733 int continue_hack_in_da_house = 0;
1736 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1741 * continue hack in da house. Ignore the ending of
1742 * this head and continue enlisting header lines.
1743 * The reason is described below.
1745 enlist(csp->headers, "");
1746 continue_hack_in_da_house = 0;
1749 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1752 * It's a bodyless continue response, don't
1753 * stop header parsing after reaching its end.
1755 * As a result Privoxy will concatenate the
1756 * next response's head and parse and deliver
1757 * the headers as if they belonged to one request.
1759 * The client will separate them because of the
1760 * empty line between them.
1762 * XXX: What we're doing here is clearly against
1763 * the intended purpose of the continue header,
1764 * and under some conditions (HTTP/1.0 client request)
1765 * it's a standard violation.
1767 * Anyway, "sort of against the spec" is preferable
1768 * to "always getting confused by Continue responses"
1769 * (Privoxy's behaviour before this hack was added)
1771 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1772 continue_hack_in_da_house = 1;
1774 else if (*header == '\0')
1777 * If the header is empty, but the Continue hack
1778 * isn't active, we can assume that we reached the
1779 * end of the buffer before we hit the end of the
1782 * Inform the caller an let it decide how to handle it.
1784 return JB_ERR_PARSE;
1787 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1790 * XXX: Should we quit the request and return a
1791 * out of memory error page instead?
1793 log_error(LOG_LEVEL_ERROR,
1794 "Out of memory while enlisting server headers. %s lost.",
1804 /*********************************************************************
1806 * Function : crunch_reason
1808 * Description : Translates the crunch reason code into a string.
1811 * 1 : rsp = a http_response
1813 * Returns : A string with the crunch reason or an error description.
1815 *********************************************************************/
1816 static const char *crunch_reason(const struct http_response *rsp)
1818 char * reason = NULL;
1820 assert(rsp != NULL);
1823 return "Internal error while searching for crunch reason";
1826 switch (rsp->reason)
1828 case RSP_REASON_UNSUPPORTED:
1829 reason = "Unsupported HTTP feature";
1831 case RSP_REASON_BLOCKED:
1834 case RSP_REASON_UNTRUSTED:
1835 reason = "Untrusted";
1837 case RSP_REASON_REDIRECTED:
1838 reason = "Redirected";
1840 case RSP_REASON_CGI_CALL:
1841 reason = "CGI Call";
1843 case RSP_REASON_NO_SUCH_DOMAIN:
1844 reason = "DNS failure";
1846 case RSP_REASON_FORWARDING_FAILED:
1847 reason = "Forwarding failed";
1849 case RSP_REASON_CONNECT_FAILED:
1850 reason = "Connection failure";
1852 case RSP_REASON_OUT_OF_MEMORY:
1853 reason = "Out of memory (may mask other reasons)";
1856 reason = "No reason recorded";
1864 /*********************************************************************
1866 * Function : send_crunch_response
1868 * Description : Delivers already prepared response for
1869 * intercepted requests, logs the interception
1870 * and frees the response.
1873 * 1 : csp = Current client state (buffers, headers, etc...)
1874 * 1 : rsp = Fully prepared response. Will be freed on exit.
1876 * Returns : Nothing.
1878 *********************************************************************/
1879 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1881 const struct http_request *http = csp->http;
1882 char status_code[4];
1884 assert(rsp != NULL);
1885 assert(rsp->head != NULL);
1890 * Not supposed to happen. If it does
1891 * anyway, treat it as an unknown error.
1893 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1894 /* return code doesn't matter */
1899 /* If rsp is still NULL, we have serious internal problems. */
1900 log_error(LOG_LEVEL_FATAL,
1901 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1905 * Extract the status code from the actual head
1906 * that was send to the client. It is the only
1907 * way to get it right for all requests, including
1908 * the fixed ones for out-of-memory problems.
1910 * A head starts like this: 'HTTP/1.1 200...'
1914 status_code[0] = rsp->head[9];
1915 status_code[1] = rsp->head[10];
1916 status_code[2] = rsp->head[11];
1917 status_code[3] = '\0';
1919 /* Write the answer to the client */
1920 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1921 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1923 /* There is nothing we can do about it. */
1924 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1927 /* Log that the request was crunched and why. */
1928 log_error(LOG_LEVEL_CRUNCH, "%s: %s", crunch_reason(rsp), http->url);
1929 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %u",
1930 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1932 /* Clean up and return */
1933 if (cgi_error_memory() != rsp)
1935 free_http_response(rsp);
1942 /*********************************************************************
1944 * Function : request_contains_null_bytes
1946 * Description : Checks for NULL bytes in the request and sends
1947 * an error message to the client if any were found.
1949 * XXX: currently not used, see comment in chat().
1952 * 1 : csp = Current client state (buffers, headers, etc...)
1953 * 2 : buf = Data from the client's request to check.
1954 * 3 : len = The data length.
1956 * Returns : TRUE if the request contained one or more NULL bytes, or
1959 *********************************************************************/
1960 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1962 size_t c_len; /* Request lenght when treated as C string */
1964 c_len = strlen(buf);
1969 * Null byte(s) found. Log the request,
1970 * return an error response and hang up.
1972 size_t tmp_len = c_len;
1977 * Replace NULL byte(s) with '°' characters
1978 * so the request can be logged as string.
1979 * XXX: Is there a better replacement character?
1982 tmp_len += strlen(buf+tmp_len);
1983 } while (tmp_len < len);
1985 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1986 "(length=%d, strlen=%u).", csp->ip_addr_str, len, c_len);
1987 log_error(LOG_LEVEL_HEADER,
1988 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1990 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1992 /* XXX: Log correct size */
1993 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2003 /*********************************************************************
2005 * Function : crunch_response_triggered
2007 * Description : Checks if the request has to be crunched,
2008 * and delivers the crunch response if necessary.
2011 * 1 : csp = Current client state (buffers, headers, etc...)
2012 * 2 : crunchers = list of cruncher functions to run
2014 * Returns : TRUE if the request was answered with a crunch response
2017 *********************************************************************/
2018 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
2020 struct http_response *rsp = NULL;
2021 const struct cruncher *c;
2024 * If CGI request crunching is disabled,
2025 * check the CGI dispatcher out of order to
2026 * prevent unintentional blocks or redirects.
2028 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
2029 && (NULL != (rsp = dispatch_cgi(csp))))
2031 /* Deliver, log and free the interception response. */
2032 send_crunch_response(csp, rsp);
2036 for (c = crunchers; c->cruncher != NULL; c++)
2039 * Check the cruncher if either Privoxy is toggled
2040 * on and the request isn't forced, or if the cruncher
2041 * applies to forced requests as well.
2043 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
2044 !(csp->flags & CSP_FLAG_FORCED)) ||
2045 (c->flags & CF_IGNORE_FORCE))
2047 rsp = c->cruncher(csp);
2050 /* Deliver, log and free the interception response. */
2051 send_crunch_response(csp, rsp);
2052 #ifdef FEATURE_STATISTICS
2053 if (c->flags & CF_COUNT_AS_REJECT)
2055 csp->flags |= CSP_FLAG_REJECTED;
2057 #endif /* def FEATURE_STATISTICS */
2068 /*********************************************************************
2070 * Function : build_request_line
2072 * Description : Builds the HTTP request line.
2074 * If a HTTP forwarder is used it expects the whole URL,
2075 * web servers only get the path.
2078 * 1 : csp = Current client state (buffers, headers, etc...)
2079 * 2 : fwd = The forwarding spec used for the request
2080 * XXX: Should use http->fwd instead.
2081 * 3 : request_line = The old request line which will be replaced.
2083 * Returns : Nothing. Terminates in case of memory problems.
2085 *********************************************************************/
2086 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
2088 struct http_request *http = csp->http;
2090 assert(http->ssl == 0);
2093 * Downgrade http version from 1.1 to 1.0
2094 * if +downgrade action applies.
2096 if ( (csp->action->flags & ACTION_DOWNGRADE)
2097 && (!strcmpic(http->ver, "HTTP/1.1")))
2100 http->ver = strdup("HTTP/1.0");
2102 if (http->ver == NULL)
2104 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
2109 * Rebuild the request line.
2111 freez(*request_line);
2112 *request_line = strdup(http->gpc);
2113 string_append(request_line, " ");
2115 if (fwd->forward_host)
2117 string_append(request_line, http->url);
2121 string_append(request_line, http->path);
2123 string_append(request_line, " ");
2124 string_append(request_line, http->ver);
2126 if (*request_line == NULL)
2128 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
2130 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
2134 /*********************************************************************
2136 * Function : change_request_destination
2138 * Description : Parse a (rewritten) request line and regenerate
2139 * the http request data.
2142 * 1 : csp = Current client state (buffers, headers, etc...)
2144 * Returns : Forwards the parse_http_request() return code.
2145 * Terminates in case of memory problems.
2147 *********************************************************************/
2148 static jb_err change_request_destination(struct client_state *csp)
2150 struct http_request *http = csp->http;
2153 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
2154 free_http_request(http);
2155 err = parse_http_request(csp->headers->first->str, http);
2156 if (JB_ERR_OK != err)
2158 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
2159 jb_err_to_string(err));
2163 /* XXX: ocmd is a misleading name */
2164 http->ocmd = strdup(http->cmd);
2165 if (http->ocmd == NULL)
2167 log_error(LOG_LEVEL_FATAL,
2168 "Out of memory copying rewritten HTTP request line");
2176 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2177 /*********************************************************************
2179 * Function : server_response_is_complete
2181 * Description : Determines whether we should stop reading
2182 * from the server socket.
2185 * 1 : csp = Current client state (buffers, headers, etc...)
2186 * 2 : content_length = Length of content received so far.
2188 * Returns : TRUE if the response is complete,
2191 *********************************************************************/
2192 static int server_response_is_complete(struct client_state *csp,
2193 unsigned long long content_length)
2195 int content_length_known = !!(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2197 if (!strcmpic(csp->http->gpc, "HEAD"))
2200 * "HEAD" implies no body, we are thus expecting
2201 * no content. XXX: incomplete "list" of methods?
2203 csp->expected_content_length = 0;
2204 content_length_known = TRUE;
2207 if (csp->http->status == 304)
2210 * Expect no body. XXX: incomplete "list" of status codes?
2212 csp->expected_content_length = 0;
2213 content_length_known = TRUE;
2216 return (content_length_known && ((0 == csp->expected_content_length)
2217 || (csp->expected_content_length <= content_length)));
2221 /*********************************************************************
2223 * Function : wait_for_alive_connections
2225 * Description : Waits for alive connections to timeout.
2231 *********************************************************************/
2232 static void wait_for_alive_connections()
2234 int connections_alive = close_unusable_connections();
2236 while (0 < connections_alive)
2238 log_error(LOG_LEVEL_CONNECT,
2239 "Waiting for %d connections to timeout.",
2242 connections_alive = close_unusable_connections();
2245 log_error(LOG_LEVEL_CONNECT, "No connections to wait for left.");
2248 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2251 /*********************************************************************
2253 * Function : mark_server_socket_tainted
2255 * Description : Makes sure we don't reuse a server socket
2256 * (if we didn't read everything the server sent
2257 * us reusing the socket would lead to garbage).
2260 * 1 : csp = Current client state (buffers, headers, etc...)
2264 *********************************************************************/
2265 static void mark_server_socket_tainted(struct client_state *csp)
2267 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
2269 log_error(LOG_LEVEL_CONNECT, "Unsetting keep-alive flag.");
2270 csp->flags &= ~CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE;
2274 /*********************************************************************
2276 * Function : get_request_line
2278 * Description : Read the client request line.
2281 * 1 : csp = Current client state (buffers, headers, etc...)
2283 * Returns : Pointer to request line or NULL in case of errors.
2285 *********************************************************************/
2286 static char *get_request_line(struct client_state *csp)
2288 char buf[BUFFER_SIZE];
2289 char *request_line = NULL;
2292 memset(buf, 0, sizeof(buf));
2296 if (!data_is_available(csp->cfd, csp->config->socket_timeout))
2298 log_error(LOG_LEVEL_ERROR,
2299 "Stopped waiting for the request line.");
2300 write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE,
2301 strlen(CONNECTION_TIMEOUT_RESPONSE));
2305 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2307 if (len <= 0) return NULL;
2310 * If there is no memory left for buffering the
2311 * request, there is nothing we can do but hang up
2313 if (add_to_iob(csp, buf, len))
2318 request_line = get_header(csp->iob);
2320 } while ((NULL != request_line) && ('\0' == *request_line));
2322 return request_line;
2327 /*********************************************************************
2329 * Function : receive_client_request
2331 * Description : Read the client's request (more precisely the
2332 * client headers) and answer it if necessary.
2334 * Note that since we're not using select() we could get
2335 * blocked here if a client connected, then didn't say
2339 * 1 : csp = Current client state (buffers, headers, etc...)
2341 * Returns : JB_ERR_OK, JB_ERR_PARSE or JB_ERR_MEMORY
2343 *********************************************************************/
2344 static jb_err receive_client_request(struct client_state *csp)
2346 char buf[BUFFER_SIZE];
2349 struct http_request *http;
2353 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2354 struct list header_list;
2355 struct list *headers = &header_list;
2359 memset(buf, 0, sizeof(buf));
2361 req = get_request_line(csp);
2364 return JB_ERR_PARSE;
2366 assert(*req != '\0');
2368 if (client_protocol_is_unsupported(csp, req))
2370 return JB_ERR_PARSE;
2373 #ifdef FEATURE_FORCE_LOAD
2375 * If this request contains the FORCE_PREFIX and blocks
2376 * aren't enforced, get rid of it and set the force flag.
2378 if (strstr(req, FORCE_PREFIX))
2380 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2382 log_error(LOG_LEVEL_FORCE,
2383 "Ignored force prefix in request: \"%s\".", req);
2387 strclean(req, FORCE_PREFIX);
2388 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2389 csp->flags |= CSP_FLAG_FORCED;
2392 #endif /* def FEATURE_FORCE_LOAD */
2394 err = parse_http_request(req, http);
2396 if (JB_ERR_OK != err)
2398 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2399 /* XXX: Use correct size */
2400 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2401 log_error(LOG_LEVEL_ERROR,
2402 "Couldn't parse request line received from %s: %s",
2403 csp->ip_addr_str, jb_err_to_string(err));
2405 free_http_request(http);
2406 return JB_ERR_PARSE;
2409 /* grab the rest of the client's headers */
2413 p = get_header(csp->iob);
2417 /* There are no additional headers to read. */
2424 * We didn't receive a complete header
2425 * line yet, get the rest of it.
2427 if (!data_is_available(csp->cfd, csp->config->socket_timeout))
2429 log_error(LOG_LEVEL_ERROR,
2430 "Stopped grabbing the client headers.");
2431 return JB_ERR_PARSE;
2434 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2437 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2438 destroy_list(headers);
2439 return JB_ERR_PARSE;
2442 if (add_to_iob(csp, buf, len))
2445 * If there is no memory left for buffering the
2446 * request, there is nothing we can do but hang up
2448 destroy_list(headers);
2449 return JB_ERR_MEMORY;
2455 * We were able to read a complete
2456 * header and can finaly enlist it.
2463 if (http->host == NULL)
2466 * If we still don't know the request destination,
2467 * the request is invalid or the client uses
2468 * Privoxy without its knowledge.
2470 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2473 * Our attempts to get the request destination
2474 * elsewhere failed or Privoxy is configured
2475 * to only accept proxy requests.
2477 * An error response has already been send
2478 * and we're done here.
2480 return JB_ERR_PARSE;
2485 * Determine the actions for this URL
2487 #ifdef FEATURE_TOGGLE
2488 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2490 /* Most compatible set of actions (i.e. none) */
2491 init_current_action(csp->action);
2494 #endif /* ndef FEATURE_TOGGLE */
2496 get_url_actions(csp, http);
2500 * Save a copy of the original request for logging
2502 http->ocmd = strdup(http->cmd);
2503 if (http->ocmd == NULL)
2505 log_error(LOG_LEVEL_FATAL,
2506 "Out of memory copying HTTP request line");
2508 enlist(csp->headers, http->cmd);
2510 /* Append the previously read headers */
2511 list_append_list_unique(csp->headers, headers);
2512 destroy_list(headers);
2519 /*********************************************************************
2521 * Function : parse_client_request
2523 * Description : Parses the client's request and decides what to do
2526 * Note that since we're not using select() we could get
2527 * blocked here if a client connected, then didn't say
2531 * 1 : csp = Current client state (buffers, headers, etc...)
2533 * Returns : JB_ERR_OK or JB_ERR_PARSE
2535 *********************************************************************/
2536 static jb_err parse_client_request(struct client_state *csp)
2538 struct http_request *http = csp->http;
2541 err = sed(csp, FILTER_CLIENT_HEADERS);
2542 if (JB_ERR_OK != err)
2544 /* XXX: Should be handled in sed(). */
2545 assert(err == JB_ERR_PARSE);
2546 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers.");
2548 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2550 /* Check request line for rewrites. */
2551 if ((NULL == csp->headers->first->str)
2552 || (strcmp(http->cmd, csp->headers->first->str) &&
2553 (JB_ERR_OK != change_request_destination(csp))))
2556 * A header filter broke the request line - bail out.
2558 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2559 /* XXX: Use correct size */
2560 log_error(LOG_LEVEL_CLF,
2561 "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2562 log_error(LOG_LEVEL_ERROR,
2563 "Invalid request line after applying header filters.");
2564 free_http_request(http);
2566 return JB_ERR_PARSE;
2574 /*********************************************************************
2578 * Description : Once a connection to the client has been accepted,
2579 * this function is called (via serve()) to handle the
2580 * main business of the communication. When this
2581 * function returns, the caller must close the client
2584 * FIXME: chat is nearly thousand lines long.
2588 * 1 : csp = Current client state (buffers, headers, etc...)
2590 * Returns : Nothing.
2592 *********************************************************************/
2593 static void chat(struct client_state *csp)
2595 char buf[BUFFER_SIZE];
2602 int ms_iis5_hack = 0;
2603 unsigned long long byte_count = 0;
2604 int forwarded_connect_retries = 0;
2605 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2606 const struct forward_spec *fwd;
2607 struct http_request *http;
2608 long len = 0; /* for buffer sizes (and negative error codes) */
2610 /* Function that does the content filtering for the current request */
2611 filter_function_ptr content_filter = NULL;
2613 /* Skeleton for HTTP response, if we should intercept the request */
2614 struct http_response *rsp;
2615 struct timeval timeout;
2617 memset(buf, 0, sizeof(buf));
2621 if (receive_client_request(csp) != JB_ERR_OK)
2625 if (parse_client_request(csp) != JB_ERR_OK)
2630 /* decide how to route the HTTP request */
2631 fwd = forward_url(csp, http);
2634 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2635 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2640 * build the http request to send to the server
2641 * we have to do one of the following:
2643 * create = use the original HTTP request to create a new
2644 * HTTP request that has either the path component
2645 * without the http://domainspec (w/path) or the
2646 * full orininal URL (w/url)
2647 * Note that the path and/or the HTTP version may
2648 * have been altered by now.
2650 * connect = Open a socket to the host:port of the server
2651 * and short-circuit server and client socket.
2653 * pass = Pass the request unchanged if forwarding a CONNECT
2654 * request to a parent proxy. Note that we'll be sending
2655 * the CFAIL message ourselves if connecting to the parent
2656 * fails, but we won't send a CSUCCEED message if it works,
2657 * since that would result in a double message (ours and the
2658 * parent's). After sending the request to the parent, we simply
2661 * here's the matrix:
2664 * +--------+--------+
2666 * 0 | create | connect|
2668 * Forwarding +--------+--------+
2670 * 1 | create | pass |
2672 * +--------+--------+
2676 if (http->ssl && connect_port_is_forbidden(csp))
2678 const char *acceptable_connect_ports =
2679 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2680 assert(NULL != acceptable_connect_ports);
2681 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2682 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2683 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2684 csp->action->flags |= ACTION_BLOCK;
2690 freez(csp->headers->first->str);
2691 build_request_line(csp, fwd, &csp->headers->first->str);
2695 * We have a request. Check if one of the crunchers wants it.
2697 if (crunch_response_triggered(csp, crunchers_all))
2700 * Yes. The client got the crunch response
2701 * and we are done here after cleaning up.
2703 /* XXX: why list_remove_all()? */
2704 list_remove_all(csp->headers);
2709 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2711 if (fwd->forward_host)
2713 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2714 fwd->forward_host, fwd->forward_port, http->hostport);
2718 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2721 /* here we connect to the server, gateway, or the forwarder */
2723 while ((csp->sfd = forwarded_connect(fwd, http, csp))
2724 && (errno == EINVAL)
2725 && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2727 log_error(LOG_LEVEL_ERROR,
2728 "failed request #%u to connect to %s. Trying again.",
2729 forwarded_connect_retries, http->hostport);
2732 if (csp->sfd == JB_INVALID_SOCKET)
2734 if (fwd->type != SOCKS_NONE)
2737 rsp = error_response(csp, "forwarding-failed", errno);
2739 else if (errno == EINVAL)
2741 rsp = error_response(csp, "no-such-domain", errno);
2745 rsp = error_response(csp, "connect-failed", errno);
2746 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2750 /* Write the answer to the client */
2753 send_crunch_response(csp, rsp);
2759 hdr = list_to_text(csp->headers);
2762 /* FIXME Should handle error properly */
2763 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2765 list_remove_all(csp->headers);
2767 if (fwd->forward_host || (http->ssl == 0))
2770 * Write the client's (modified) header to the server
2771 * (along with anything else that may be in the buffer)
2773 if (write_socket(csp->sfd, hdr, strlen(hdr))
2774 || (flush_socket(csp->sfd, csp->iob) < 0))
2776 log_error(LOG_LEVEL_CONNECT,
2777 "write header to: %s failed: %E", http->hostport);
2779 rsp = error_response(csp, "connect-failed", errno);
2782 send_crunch_response(csp, rsp);
2792 * We're running an SSL tunnel and we're not forwarding,
2793 * so just send the "connect succeeded" message to the
2794 * client, flush the rest, and get out of the way.
2796 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2804 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2806 /* we're finished with the client's header */
2809 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2811 /* pass data between the client and server
2812 * until one or the other shuts down the connection.
2821 * FD_ZERO here seems to point to an errant macro which crashes.
2822 * So do this by hand for now...
2824 memset(&rfds,0x00,sizeof(fd_set));
2828 FD_SET(csp->cfd, &rfds);
2829 FD_SET(csp->sfd, &rfds);
2831 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2832 if ((csp->flags & CSP_FLAG_CHUNKED)
2833 && !(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
2834 && ((csp->iob->eod - csp->iob->cur) >= 5)
2835 && !memcmp(csp->iob->eod-5, "0\r\n\r\n", 5))
2837 log_error(LOG_LEVEL_CONNECT,
2838 "Looks like we read the last chunk together with "
2839 "the server headers. We better stop reading.");
2840 byte_count = (unsigned long long)(csp->iob->eod - csp->iob->cur);
2841 csp->expected_content_length = byte_count;
2842 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2844 if (server_body && server_response_is_complete(csp, byte_count))
2846 log_error(LOG_LEVEL_CONNECT,
2847 "Done reading from server. Expected content length: %llu. "
2848 "Actual content length: %llu. Most recently received: %d.",
2849 csp->expected_content_length, byte_count, len);
2852 * XXX: should not jump around,
2853 * chat() is complicated enough already.
2857 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2859 timeout.tv_sec = csp->config->socket_timeout;
2860 timeout.tv_usec = 0;
2861 n = select((int)maxfd+1, &rfds, NULL, NULL, &timeout);
2865 log_error(LOG_LEVEL_ERROR,
2866 "Didn't receive data in time: %s", http->url);
2867 if ((byte_count == 0) && (http->ssl == 0))
2869 write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE,
2870 strlen(CONNECTION_TIMEOUT_RESPONSE));
2872 mark_server_socket_tainted(csp);
2877 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2878 mark_server_socket_tainted(csp);
2883 * This is the body of the browser's request,
2884 * just read and write it.
2886 if (FD_ISSET(csp->cfd, &rfds))
2888 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2892 /* XXX: not sure if this is necessary. */
2893 mark_server_socket_tainted(csp);
2894 break; /* "game over, man" */
2897 if (write_socket(csp->sfd, buf, (size_t)len))
2899 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2900 mark_server_socket_tainted(csp);
2907 * The server wants to talk. It could be the header or the body.
2908 * If `hdr' is null, then it's the header otherwise it's the body.
2909 * FIXME: Does `hdr' really mean `host'? No.
2911 if (FD_ISSET(csp->sfd, &rfds))
2914 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2918 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2920 if (http->ssl && (fwd->forward_host == NULL))
2923 * Just hang up. We already confirmed the client's CONNECT
2924 * request with status code 200 and unencrypted content is
2925 * no longer welcome.
2927 log_error(LOG_LEVEL_ERROR,
2928 "CONNECT already confirmed. Unable to tell the client about the problem.");
2931 else if (byte_count)
2934 * Just hang up. We already transmitted the original headers
2935 * and parts of the original content and therefore missed the
2936 * chance to send an error message (without risking data corruption).
2938 * XXX: we could retry with a fancy range request here.
2940 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2941 "Unable to tell the client about the problem.");
2942 mark_server_socket_tainted(csp);
2946 rsp = error_response(csp, "connect-failed", errno);
2949 send_crunch_response(csp, rsp);
2955 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2956 if (csp->flags & CSP_FLAG_CHUNKED)
2958 if ((len >= 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
2960 /* XXX: this is a temporary hack */
2961 log_error(LOG_LEVEL_CONNECT,
2962 "Looks like we reached the end of the last chunk. "
2963 "We better stop reading.");
2964 csp->expected_content_length = byte_count + (unsigned long long)len;
2965 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2969 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2972 * Add a trailing zero to let be able to use string operations.
2973 * XXX: do we still need this with filter_popups gone?
2978 * Normally, this would indicate that we've read
2979 * as much as the server has sent us and we can
2980 * close the client connection. However, Microsoft
2981 * in its wisdom has released IIS/5 with a bug that
2982 * prevents it from sending the trailing \r\n in
2983 * a 302 redirect header (and possibly other headers).
2984 * To work around this if we've haven't parsed
2985 * a full header we'll append a trailing \r\n
2986 * and see if this now generates a valid one.
2988 * This hack shouldn't have any impacts. If we've
2989 * already transmitted the header or if this is a
2990 * SSL connection, then we won't bother with this
2991 * hack. So we only work on partially received
2992 * headers. If we append a \r\n and this still
2993 * doesn't generate a valid header, then we won't
2994 * transmit anything to the client.
2999 if (server_body || http->ssl)
3002 * If we have been buffering up the document,
3003 * now is the time to apply content modification
3004 * and send the result to the client.
3008 p = execute_content_filter(csp, content_filter);
3010 * If the content filter fails, use the original
3011 * buffer and length.
3012 * (see p != NULL ? p : csp->iob->cur below)
3016 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
3019 if (JB_ERR_OK != update_server_headers(csp))
3021 log_error(LOG_LEVEL_FATAL,
3022 "Failed to update server headers. after filtering.");
3025 hdr = list_to_text(csp->headers);
3028 /* FIXME Should handle error properly */
3029 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3032 if (write_socket(csp->cfd, hdr, strlen(hdr))
3033 || write_socket(csp->cfd,
3034 ((p != NULL) ? p : csp->iob->cur), (size_t)csp->content_length))
3036 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
3039 mark_server_socket_tainted(csp);
3047 break; /* "game over, man" */
3051 * This is NOT the body, so
3052 * Let's pretend the server just sent us a blank line.
3054 snprintf(buf, sizeof(buf), "\r\n");
3055 len = (int)strlen(buf);
3058 * Now, let the normal header parsing algorithm below do its
3059 * job. If it fails, we'll exit instead of continuing.
3066 * If this is an SSL connection or we're in the body
3067 * of the server document, just write it to the client,
3068 * unless we need to buffer the body for later content-filtering
3070 if (server_body || http->ssl)
3075 * If there is no memory left for buffering the content, or the buffer limit
3076 * has been reached, switch to non-filtering mode, i.e. make & write the
3077 * header, flush the iob and buf, and get out of the way.
3079 if (add_to_iob(csp, buf, len))
3084 log_error(LOG_LEVEL_INFO,
3085 "Flushing header and buffers. Stepping back from filtering.");
3087 hdr = list_to_text(csp->headers);
3091 * Memory is too tight to even generate the header.
3092 * Send our static "Out-of-memory" page.
3094 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
3095 rsp = cgi_error_memory();
3096 send_crunch_response(csp, rsp);
3097 mark_server_socket_tainted(csp);
3100 hdrlen = strlen(hdr);
3102 if (write_socket(csp->cfd, hdr, hdrlen)
3103 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
3104 || (write_socket(csp->cfd, buf, (size_t)len)))
3106 log_error(LOG_LEVEL_CONNECT,
3107 "Flush header and buffers to client failed: %E");
3109 mark_server_socket_tainted(csp);
3114 * Reset the byte_count to the amount of bytes
3115 * we just flushed. len will be added a few lines below,
3116 * hdrlen doesn't matter for LOG_LEVEL_CLF.
3118 byte_count = (unsigned long long)flushed;
3120 content_filter = NULL;
3126 if (write_socket(csp->cfd, buf, (size_t)len))
3128 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
3129 mark_server_socket_tainted(csp);
3133 byte_count += (unsigned long long)len;
3138 const char *header_start;
3140 * We're still looking for the end of the server's header.
3141 * Buffer up the data we just read. If that fails, there's
3142 * little we can do but send our static out-of-memory page.
3144 if (add_to_iob(csp, buf, len))
3146 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
3147 rsp = cgi_error_memory();
3148 send_crunch_response(csp, rsp);
3149 mark_server_socket_tainted(csp);
3153 header_start = csp->iob->cur;
3155 /* Convert iob into something sed() can digest */
3156 if (JB_ERR_PARSE == get_server_headers(csp))
3161 * Well, we tried our MS IIS/5 hack and it didn't work.
3162 * The header is incomplete and there isn't anything
3163 * we can do about it.
3165 log_error(LOG_LEVEL_INFO,
3166 "MS IIS5 hack didn't produce valid headers.");
3172 * Since we have to wait for more from the server before
3173 * we can parse the headers we just continue here.
3175 long header_offset = csp->iob->cur - header_start;
3176 assert(csp->iob->cur >= header_start);
3177 byte_count += (unsigned long long)(len - header_offset);
3178 log_error(LOG_LEVEL_CONNECT, "Continuing buffering headers. "
3179 "byte_count: %llu. header_offset: %d. len: %d.",
3180 byte_count, header_offset, len);
3185 /* Did we actually get anything? */
3186 if (NULL == csp->headers->first)
3188 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
3189 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3190 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
3191 free_http_request(http);
3192 mark_server_socket_tainted(csp);
3196 assert(csp->headers->first->str);
3198 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
3199 strncmpic(csp->headers->first->str, "ICY", 3))
3202 * It doesn't look like a HTTP (or Shoutcast) response:
3203 * tell the client and log the problem.
3205 if (strlen(csp->headers->first->str) > 30)
3207 csp->headers->first->str[30] = '\0';
3209 log_error(LOG_LEVEL_ERROR,
3210 "Invalid server or forwarder response. Starts with: %s",
3211 csp->headers->first->str);
3212 log_error(LOG_LEVEL_CLF,
3213 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3214 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
3215 strlen(INVALID_SERVER_HEADERS_RESPONSE));
3216 free_http_request(http);
3217 mark_server_socket_tainted(csp);
3222 * We have now received the entire server header,
3223 * filter it and send the result to the client
3225 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
3227 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
3229 hdr = list_to_text(csp->headers);
3232 /* FIXME Should handle error properly */
3233 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3236 if (crunch_response_triggered(csp, crunchers_light))
3239 * One of the tags created by a server-header
3240 * tagger triggered a crunch. We already
3241 * delivered the crunch response to the client
3242 * and are done here after cleaning up.
3245 mark_server_socket_tainted(csp);
3248 /* Buffer and pcrs filter this if appropriate. */
3250 if (!http->ssl) /* We talk plaintext */
3252 content_filter = get_filter_function(csp);
3255 * Only write if we're not buffering for content modification
3257 if (!content_filter)
3260 * Write the server's (modified) header to
3261 * the client (along with anything else that
3262 * may be in the buffer)
3265 if (write_socket(csp->cfd, hdr, strlen(hdr))
3266 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
3268 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
3271 * The write failed, so don't bother mentioning it
3272 * to the client... it probably can't hear us anyway.
3275 mark_server_socket_tainted(csp);
3279 byte_count += (unsigned long long)len;
3284 * XXX: the header lenght should probably
3285 * be calculated by get_server_headers().
3287 long header_length = csp->iob->cur - header_start;
3288 assert(csp->iob->cur > header_start);
3289 byte_count += (unsigned long long)(len - header_length);
3292 /* we're finished with the server's header */
3298 * If this was a MS IIS/5 hack then it means the server
3299 * has already closed the connection. Nothing more to read.
3304 log_error(LOG_LEVEL_INFO,
3305 "Closed server connection detected with MS IIS5 hack enabled.");
3311 mark_server_socket_tainted(csp);
3312 return; /* huh? we should never get here */
3315 if (csp->content_length == 0)
3318 * If Privoxy didn't recalculate the Content-Lenght,
3319 * byte_count is still correct.
3321 csp->content_length = byte_count;
3324 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3325 if ((csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
3326 && (csp->expected_content_length != byte_count))
3328 log_error(LOG_LEVEL_CONNECT,
3329 "Received %llu bytes while expecting %llu.",
3330 byte_count, csp->expected_content_length);
3331 mark_server_socket_tainted(csp);
3335 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %llu",
3336 csp->ip_addr_str, http->ocmd, csp->content_length);
3340 /*********************************************************************
3344 * Description : This is little more than chat. We only "serve" to
3345 * to close (or remember) any socket that chat may have
3349 * 1 : csp = Current client state (buffers, headers, etc...)
3353 *********************************************************************/
3355 void serve(struct client_state *csp)
3356 #else /* ifndef AMIGA */
3357 static void serve(struct client_state *csp)
3358 #endif /* def AMIGA */
3362 if (csp->sfd != JB_INVALID_SOCKET)
3364 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3365 static int monitor_thread_running = 0;
3367 if ((csp->config->feature_flags & RUNTIME_FEATURE_CONNECTION_KEEP_ALIVE)
3368 && (csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
3370 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
3371 close_socket(csp->cfd);
3372 csp->cfd = JB_INVALID_SOCKET;
3373 privoxy_mutex_lock(&connection_reuse_mutex);
3374 if (!monitor_thread_running)
3376 monitor_thread_running = 1;
3377 privoxy_mutex_unlock(&connection_reuse_mutex);
3378 wait_for_alive_connections();
3379 privoxy_mutex_lock(&connection_reuse_mutex);
3380 monitor_thread_running = 0;
3382 privoxy_mutex_unlock(&connection_reuse_mutex);
3386 forget_connection(csp->sfd);
3387 close_socket(csp->sfd);
3390 close_socket(csp->sfd);
3391 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3394 if (csp->cfd != JB_INVALID_SOCKET)
3396 close_socket(csp->cfd);
3399 csp->flags &= ~CSP_FLAG_ACTIVE;
3405 /*********************************************************************
3407 * Function : server_thread
3409 * Description : We only exist to call `serve' in a threaded environment.
3412 * 1 : data = Current client state (buffers, headers, etc...)
3414 * Returns : Always 0.
3416 *********************************************************************/
3417 static int32 server_thread(void *data)
3419 serve((struct client_state *) data);
3426 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3427 /*********************************************************************
3431 * Description : Print usage info & exit.
3433 * Parameters : Pointer to argv[0] for identifying ourselves
3437 *********************************************************************/
3438 static void usage(const char *myname)
3440 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3444 #endif /* defined(unix) */
3447 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3448 #endif /* defined(unix) */
3449 "[--version] [configfile]\n"
3450 "Aborting\n", myname);
3455 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3458 #ifdef MUTEX_LOCKS_AVAILABLE
3459 /*********************************************************************
3461 * Function : privoxy_mutex_lock
3463 * Description : Locks a mutex.
3466 * 1 : mutex = The mutex to lock.
3468 * Returns : Void. May exit in case of errors.
3470 *********************************************************************/
3471 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3473 #ifdef FEATURE_PTHREAD
3474 int err = pthread_mutex_lock(mutex);
3477 if (mutex != &log_mutex)
3479 log_error(LOG_LEVEL_FATAL,
3480 "Mutex locking failed: %s.\n", strerror(err));
3485 EnterCriticalSection(mutex);
3486 #endif /* def FEATURE_PTHREAD */
3490 /*********************************************************************
3492 * Function : privoxy_mutex_unlock
3494 * Description : Unlocks a mutex.
3497 * 1 : mutex = The mutex to unlock.
3499 * Returns : Void. May exit in case of errors.
3501 *********************************************************************/
3502 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3504 #ifdef FEATURE_PTHREAD
3505 int err = pthread_mutex_unlock(mutex);
3508 if (mutex != &log_mutex)
3510 log_error(LOG_LEVEL_FATAL,
3511 "Mutex unlocking failed: %s.\n", strerror(err));
3516 LeaveCriticalSection(mutex);
3517 #endif /* def FEATURE_PTHREAD */
3521 /*********************************************************************
3523 * Function : privoxy_mutex_init
3525 * Description : Prepares a mutex.
3528 * 1 : mutex = The mutex to initialize.
3530 * Returns : Void. May exit in case of errors.
3532 *********************************************************************/
3533 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3535 #ifdef FEATURE_PTHREAD
3536 int err = pthread_mutex_init(mutex, 0);
3539 printf("Fatal error. Mutex initialization failed: %s.\n",
3544 InitializeCriticalSection(mutex);
3545 #endif /* def FEATURE_PTHREAD */
3547 #endif /* def MUTEX_LOCKS_AVAILABLE */
3549 /*********************************************************************
3551 * Function : initialize_mutexes
3553 * Description : Prepares mutexes if mutex support is available.
3557 * Returns : Void, exits in case of errors.
3559 *********************************************************************/
3560 static void initialize_mutexes(void)
3562 #ifdef MUTEX_LOCKS_AVAILABLE
3564 * Prepare global mutex semaphores
3566 privoxy_mutex_init(&log_mutex);
3567 privoxy_mutex_init(&log_init_mutex);
3568 privoxy_mutex_init(&connection_reuse_mutex);
3571 * XXX: The assumptions below are a bit naive
3572 * and can cause locks that aren't necessary.
3574 * For example older FreeBSD versions (< 6.x?)
3575 * have no gethostbyname_r, but gethostbyname is
3578 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3579 privoxy_mutex_init(&resolver_mutex);
3580 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3582 * XXX: should we use a single mutex for
3583 * localtime() and gmtime() as well?
3585 #ifndef HAVE_GMTIME_R
3586 privoxy_mutex_init(&gmtime_mutex);
3587 #endif /* ndef HAVE_GMTIME_R */
3589 #ifndef HAVE_LOCALTIME_R
3590 privoxy_mutex_init(&localtime_mutex);
3591 #endif /* ndef HAVE_GMTIME_R */
3594 privoxy_mutex_init(&rand_mutex);
3595 #endif /* ndef HAVE_RANDOM */
3596 #endif /* def MUTEX_LOCKS_AVAILABLE */
3600 /*********************************************************************
3604 * Description : Load the config file and start the listen loop.
3605 * This function is a lot more *sane* with the `load_config'
3606 * and `listen_loop' functions; although it stills does
3607 * a *little* too much for my taste.
3610 * 1 : argc = Number of parameters (including $0).
3611 * 2 : argv = Array of (char *)'s to the parameters.
3613 * Returns : 1 if : can't open config file, unrecognized directive,
3614 * stats requested in multi-thread mode, can't open the
3615 * log file, can't open the jar file, listen port is invalid,
3616 * any load fails, and can't bind port.
3618 * Else main never returns, the process must be signaled
3619 * to terminate execution. Or, on Windows, use the
3620 * "File", "Exit" menu option.
3622 *********************************************************************/
3624 int real_main(int argc, const char *argv[])
3626 int main(int argc, const char *argv[])
3630 unsigned int random_seed;
3632 struct passwd *pw = NULL;
3633 struct group *grp = NULL;
3636 char *pre_chroot_nslookup_to_load_resolver = NULL;
3643 #if !defined(_WIN32)
3650 /* Prepare mutexes if supported and necessary. */
3651 initialize_mutexes();
3653 /* Enable logging until further notice. */
3657 * Parse the command line arguments
3659 * XXX: simply printing usage information in case of
3660 * invalid arguments isn't particularly user friendly.
3662 while (++argc_pos < argc)
3665 /* Check to see if the service must be installed or uninstalled */
3666 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3668 const char *pName = argv[argc_pos] + 9;
3671 exit( (install_service(pName)) ? 0 : 1 );
3673 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3675 const char *pName = argv[argc_pos] + 11;
3678 exit((uninstall_service(pName)) ? 0 : 1);
3680 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3682 bRunAsService = TRUE;
3683 w32_set_service_cwd();
3684 atexit(w32_service_exit_notify);
3687 #endif /* defined(_WIN32) */
3690 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3692 if (strcmp(argv[argc_pos], "--help") == 0)
3697 else if(strcmp(argv[argc_pos], "--version") == 0)
3699 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3705 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3707 set_debug_level(LOG_LEVEL_FATAL | LOG_LEVEL_ERROR | LOG_LEVEL_INFO);
3711 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3713 if (++argc_pos == argc) usage(argv[0]);
3714 pidfile = strdup(argv[argc_pos]);
3717 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3719 if (++argc_pos == argc) usage(argv[argc_pos]);
3721 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3724 if (NULL == (grp = getgrnam(p)))
3726 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3730 if (NULL == (pw = getpwnam(argv[argc_pos])))
3732 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3735 if (p != NULL) *--p = '\0';
3738 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3740 if (++argc_pos == argc) usage(argv[0]);
3741 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3744 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3748 #endif /* defined(unix) */
3750 else if (argc_pos + 1 != argc)
3753 * This is neither the last command line
3754 * option, nor was it recognized before,
3755 * therefore it must be invalid.
3761 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3763 configfile = argv[argc_pos];
3766 } /* -END- while (more arguments) */
3768 show_version(Argv[0]);
3771 if ( *configfile != '/' )
3773 char cwd[BUFFER_SIZE];
3775 size_t abs_file_size;
3777 /* make config-filename absolute here */
3778 if (NULL == getcwd(cwd, sizeof(cwd)))
3780 perror("failed to get current working directory");
3785 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3786 basedir = strdup(cwd);
3788 if (NULL == basedir ||
3789 NULL == (abs_file = malloc(abs_file_size)))
3791 perror("malloc failed");
3794 strlcpy(abs_file, basedir, abs_file_size);
3795 strlcat(abs_file, "/", abs_file_size );
3796 strlcat(abs_file, configfile, abs_file_size);
3797 configfile = abs_file;
3799 #endif /* defined unix */
3803 clients->next = NULL;
3805 /* XXX: factor out initialising after the next stable release. */
3808 #elif defined(_WIN32)
3812 random_seed = (unsigned int)time(NULL);
3814 srandom(random_seed);
3817 #endif /* ifdef HAVE_RANDOM */
3820 * Unix signal handling
3822 * Catch the abort, interrupt and terminate signals for a graceful exit
3823 * Catch the hangup signal so the errlog can be reopened.
3824 * Ignore the broken pipe signals (FIXME: Why?)
3826 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3829 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3830 const int ignored_signals[] = { SIGPIPE, 0 };
3832 for (idx = 0; catched_signals[idx] != 0; idx++)
3834 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3835 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3837 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3838 #endif /* ifdef sun */
3840 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3844 for (idx = 0; ignored_signals[idx] != 0; idx++)
3846 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3848 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3853 #else /* ifdef _WIN32 */
3854 # ifdef _WIN_CONSOLE
3856 * We *are* in a windows console app.
3857 * Print a verbose messages about FAQ's and such
3859 printf("%s", win32_blurb);
3860 # endif /* def _WIN_CONSOLE */
3861 #endif /* def _WIN32 */
3864 /* Initialize the CGI subsystem */
3865 cgi_init_error_messages();
3868 * If runnig on unix and without the --nodaemon
3869 * option, become a daemon. I.e. fork, detach
3870 * from tty and get process group leadership
3883 if ( pid < 0 ) /* error */
3888 else if ( pid != 0 ) /* parent */
3893 * must check for errors
3894 * child died due to missing files aso
3897 wpid = waitpid( pid, &status, WNOHANG );
3906 /* Should be more portable, but not as well tested */
3911 #else /* ndef __FreeBSD__ */
3913 #endif /* ndef __FreeBSD__ */
3914 fd = open("/dev/tty", O_RDONLY);
3917 /* no error check here */
3918 ioctl( fd, TIOCNOTTY,0 );
3923 * stderr (fd 2) will be closed later on,
3924 * when the config file has been parsed.
3931 } /* -END- if (!no_daemon) */
3934 * As soon as we have written the PID file, we can switch
3935 * to the user and group ID indicated by the --user option
3941 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3943 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3947 if (setgroups(1, &grp->gr_gid))
3949 log_error(LOG_LEVEL_FATAL, "setgroups() failed: %E");
3952 else if (initgroups(pw->pw_name, pw->pw_gid))
3954 log_error(LOG_LEVEL_FATAL, "initgroups() failed: %E");
3960 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3962 /* Read the time zone file from /etc before doing chroot. */
3964 if (NULL != pre_chroot_nslookup_to_load_resolver
3965 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3967 /* Initialize resolver library. */
3968 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3970 if (chroot(pw->pw_dir) < 0)
3972 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3976 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3979 if (setuid(pw->pw_uid))
3981 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3985 char putenv_dummy[64];
3987 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3988 if (putenv(putenv_dummy) != 0)
3990 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3993 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3994 if (putenv(putenv_dummy) != 0)
3996 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
4002 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
4005 #endif /* defined unix */
4008 /* This will be FALSE unless the command line specified --service
4012 /* Yup, so now we must attempt to establish a connection
4013 * with the service dispatcher. This will only work if this
4014 * process was launched by the service control manager to
4015 * actually run as a service. If this isn't the case, i've
4016 * known it take around 30 seconds or so for the call to return.
4019 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
4020 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
4022 /* Service has run, and at this point is now being stopped, so just return */
4027 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
4029 /* An error occurred. Usually it's because --service was wrongly specified
4030 * and we were unable to connect to the Service Control Dispatcher because
4031 * it wasn't expecting us and is therefore not listening.
4033 * For now, just continue below to call the listen_loop function.
4036 #endif /* def _WIN32 */
4046 /*********************************************************************
4048 * Function : bind_port_helper
4050 * Description : Bind the listen port. Handles logging, and aborts
4054 * 1 : config = Privoxy configuration. Specifies port
4057 * Returns : Port that was opened.
4059 *********************************************************************/
4060 static jb_socket bind_port_helper(struct configuration_spec * config)
4065 if (config->haddr == NULL)
4067 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
4072 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
4073 config->hport, config->haddr);
4076 result = bind_port(config->haddr, config->hport, &bfd);
4083 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
4084 "There may be another Privoxy or some other "
4085 "proxy running on port %d",
4086 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
4087 config->hport, config->hport);
4090 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
4091 "The hostname is not resolvable",
4092 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
4095 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
4096 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
4099 /* shouldn't get here */
4100 return JB_INVALID_SOCKET;
4103 config->need_bind = 0;
4110 /* Without this simple workaround we get this compiler warning from _beginthread
4111 * warning C4028: formal parameter 1 different from declaration
4113 void w32_service_listen_loop(void *p)
4117 #endif /* def _WIN32 */
4120 /*********************************************************************
4122 * Function : listen_loop
4124 * Description : bind the listen port and enter a "FOREVER" listening loop.
4130 *********************************************************************/
4131 static void listen_loop(void)
4133 struct client_state *csp = NULL;
4135 struct configuration_spec * config;
4137 config = load_config();
4139 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
4141 * XXX: Should be relocated once it no
4142 * longer needs to emit log messages.
4144 initialize_reusable_connections();
4145 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
4147 bfd = bind_port_helper(config);
4149 #ifdef FEATURE_GRACEFUL_TERMINATION
4150 while (!g_terminate)
4155 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
4156 while (waitpid(-1, NULL, WNOHANG) > 0)
4158 /* zombie children */
4160 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
4163 * Free data that was used by died threads
4169 * Re-open the errlog after HUP signal
4171 if (received_hup_signal)
4173 if (NULL != config->logfile)
4175 init_error_log(Argv[0], config->logfile);
4177 received_hup_signal = 0;
4181 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
4183 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
4187 csp->flags |= CSP_FLAG_ACTIVE;
4188 csp->sfd = JB_INVALID_SOCKET;
4190 csp->config = config = load_config();
4192 if ( config->need_bind )
4195 * Since we were listening to the "old port", we will not see
4196 * a "listen" param change until the next IJB request. So, at
4197 * least 1 more request must be made for us to find the new
4198 * setting. I am simply closing the old socket and binding the
4201 * Which-ever is correct, we will serve 1 more page via the
4202 * old settings. This should probably be a "show-proxy-args"
4203 * request. This should not be a so common of an operation
4204 * that this will hurt people's feelings.
4209 bfd = bind_port_helper(config);
4212 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
4214 if (!accept_connection(csp, bfd))
4216 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
4229 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
4232 #ifdef FEATURE_TOGGLE
4233 if (global_toggle_state)
4234 #endif /* def FEATURE_TOGGLE */
4236 csp->flags |= CSP_FLAG_TOGGLED_ON;
4239 if (run_loader(csp))
4241 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
4242 /* Never get here - LOG_LEVEL_FATAL causes program exit */
4246 if (block_acl(NULL,csp))
4248 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
4249 close_socket(csp->cfd);
4253 #endif /* def FEATURE_ACL */
4255 /* add it to the list of clients */
4256 csp->next = clients->next;
4257 clients->next = csp;
4259 if (config->multi_threaded)
4263 /* this is a switch () statment in the C preprocessor - ugh */
4264 #undef SELECTED_ONE_OPTION
4266 /* Use Pthreads in preference to native code */
4267 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
4268 #define SELECTED_ONE_OPTION
4270 pthread_t the_thread;
4271 pthread_attr_t attrs;
4273 pthread_attr_init(&attrs);
4274 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
4275 errno = pthread_create(&the_thread, &attrs,
4276 (void * (*)(void *))serve, csp);
4277 child_id = errno ? -1 : 0;
4278 pthread_attr_destroy(&attrs);
4282 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
4283 #define SELECTED_ONE_OPTION
4284 child_id = _beginthread(
4285 (void (*)(void *))serve,
4290 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
4291 #define SELECTED_ONE_OPTION
4292 child_id = _beginthread(
4293 (void(* _Optlink)(void*))serve,
4299 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
4300 #define SELECTED_ONE_OPTION
4302 thread_id tid = spawn_thread
4303 (server_thread, "server", B_NORMAL_PRIORITY, csp);
4305 if ((tid >= 0) && (resume_thread(tid) == B_OK))
4307 child_id = (int) tid;
4316 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
4317 #define SELECTED_ONE_OPTION
4318 csp->cfd = ReleaseSocket(csp->cfd, -1);
4321 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
4322 NP_Output, Output(),
4323 NP_CloseOutput, FALSE,
4324 NP_Name, (ULONG)"privoxy child",
4328 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
4329 NP_Output, Output(),
4330 NP_CloseOutput, FALSE,
4331 NP_Name, (ULONG)"privoxy child",
4332 NP_StackSize, 200*1024,
4338 ((struct Task *)child_id)->tc_UserData = csp;
4339 Signal((struct Task *)child_id, SIGF_SINGLE);
4344 #if !defined(SELECTED_ONE_OPTION)
4347 /* This block is only needed when using fork().
4348 * When using threads, the server thread was
4349 * created and run by the call to _beginthread().
4351 if (child_id == 0) /* child */
4354 #ifdef FEATURE_TOGGLE
4355 int inherited_toggle_state = global_toggle_state;
4356 #endif /* def FEATURE_TOGGLE */
4361 * If we've been toggled or we've blocked the request, tell Mom
4364 #ifdef FEATURE_TOGGLE
4365 if (inherited_toggle_state != global_toggle_state)
4367 rc |= RC_FLAG_TOGGLED;
4369 #endif /* def FEATURE_TOGGLE */
4371 #ifdef FEATURE_STATISTICS
4372 if (csp->flags & CSP_FLAG_REJECTED)
4374 rc |= RC_FLAG_BLOCKED;
4376 #endif /* ndef FEATURE_STATISTICS */
4380 else if (child_id > 0) /* parent */
4382 /* in a fork()'d environment, the parent's
4383 * copy of the client socket and the CSP
4387 #if !defined(_WIN32) && !defined(__CYGWIN__)
4389 wait( &child_status );
4392 * Evaluate child's return code: If the child has
4393 * - been toggled, toggle ourselves
4394 * - blocked its request, bump up the stats counter
4397 #ifdef FEATURE_TOGGLE
4398 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
4400 global_toggle_state = !global_toggle_state;
4402 #endif /* def FEATURE_TOGGLE */
4404 #ifdef FEATURE_STATISTICS
4406 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4410 #endif /* def FEATURE_STATISTICS */
4412 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4413 close_socket(csp->cfd);
4414 csp->flags &= ~CSP_FLAG_ACTIVE;
4418 #undef SELECTED_ONE_OPTION
4419 /* end of cpp switch () */
4421 if (child_id < 0) /* failed */
4423 char buf[BUFFER_SIZE];
4425 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
4427 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
4429 write_socket(csp->cfd, buf, strlen(buf));
4430 close_socket(csp->cfd);
4431 csp->flags &= ~CSP_FLAG_ACTIVE;
4442 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4444 /* Clean up. Aim: free all memory (no leaks) */
4445 #ifdef FEATURE_GRACEFUL_TERMINATION
4447 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4449 unload_current_config_file();
4450 unload_current_actions_file();
4451 unload_current_re_filterfile();
4452 #ifdef FEATURE_TRUST
4453 unload_current_trust_file();
4456 if (config->multi_threaded)
4463 } while ((clients->next != NULL) && (--i > 0));
4467 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4478 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4479 /* Cleanup - remove taskbar icon etc. */
4484 #endif /* FEATURE_GRACEFUL_TERMINATION */