1 const char filters_rcs[] = "$Id: filters.c,v 1.90 2007/09/02 12:44:17 fabiankeil Exp $";
2 /*********************************************************************
4 * File : $Source: /cvsroot/ijbswa/current/filters.c,v $
6 * Purpose : Declares functions to parse/crunch headers and pages.
7 * Functions declared include:
8 * `acl_addr', `add_stats', `block_acl', `block_imageurl',
9 * `block_url', `url_actions', `domain_split',
10 * `filter_popups', `forward_url', 'redirect_url',
11 * `ij_untrusted_url', `intercept_url', `pcrs_filter_respose',
12 * `ijb_send_banner', `trust_url', `gif_deanimate_response',
13 * `jpeg_inspect_response', `execute_single_pcrs_command',
14 * `rewrite_url', `get_last_url'
16 * Copyright : Written by and Copyright (C) 2001, 2004-2007 the SourceForge
17 * Privoxy team. http://www.privoxy.org/
19 * Based on the Internet Junkbuster originally written
20 * by and Copyright (C) 1997 Anonymous Coders and
21 * Junkbusters Corporation. http://www.junkbusters.com
23 * This program is free software; you can redistribute it
24 * and/or modify it under the terms of the GNU General
25 * Public License as published by the Free Software
26 * Foundation; either version 2 of the License, or (at
27 * your option) any later version.
29 * This program is distributed in the hope that it will
30 * be useful, but WITHOUT ANY WARRANTY; without even the
31 * implied warranty of MERCHANTABILITY or FITNESS FOR A
32 * PARTICULAR PURPOSE. See the GNU General Public
33 * License for more details.
35 * The GNU General Public License should be included with
36 * this file. If not, you can view it at
37 * http://www.gnu.org/copyleft/gpl.html
38 * or write to the Free Software Foundation, Inc., 59
39 * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
43 * Revision 1.90 2007/09/02 12:44:17 fabiankeil
44 * Remove newline at the end of a log_error() message.
46 * Revision 1.89 2007/08/05 13:42:23 fabiankeil
47 * #1763173 from Stefan Huehner: declare some more functions static.
49 * Revision 1.88 2007/06/01 16:41:11 fabiankeil
50 * Add forward-override{} to change the forwarding settings through
51 * action sections. This is mainly interesting to forward different
52 * clients differently (for example based on User-Agent or request
55 * Revision 1.87 2007/04/30 15:53:10 fabiankeil
56 * Make sure filters with dynamic jobs actually use them.
58 * Revision 1.86 2007/04/30 15:03:28 fabiankeil
59 * - Introduce dynamic pcrs jobs that can resolve variables.
60 * - Don't run redirect functions more than once,
61 * unless they are activated more than once.
63 * Revision 1.85 2007/03/21 12:24:47 fabiankeil
64 * - Log the content size after decompression in decompress_iob()
65 * instead of pcrs_filter_response().
67 * Revision 1.84 2007/03/20 15:16:34 fabiankeil
68 * Use dedicated header filter actions instead of abusing "filter".
69 * Replace "filter-client-headers" and "filter-client-headers"
70 * with "server-header-filter" and "client-header-filter".
72 * Revision 1.83 2007/03/17 15:20:05 fabiankeil
73 * New config option: enforce-blocks.
75 * Revision 1.82 2007/03/13 11:28:43 fabiankeil
76 * - Fix port handling in acl_addr() and use a temporary acl spec
77 * copy so error messages don't contain a truncated version.
78 * - Log size of iob before and after decompression.
80 * Revision 1.81 2007/03/05 14:40:53 fabiankeil
81 * - Cosmetical changes for LOG_LEVEL_RE_FILTER messages.
82 * - Hide the "Go there anyway" link for blocked CONNECT
83 * requests where going there anyway doesn't work anyway.
85 * Revision 1.80 2007/02/07 10:55:20 fabiankeil
86 * - Save the reason for generating http_responses.
87 * - Block (+block) with status code 403 instead of 404.
88 * - Use a different kludge to remember a failed decompression.
90 * Revision 1.79 2007/01/31 16:21:38 fabiankeil
91 * Search for Max-Forwards headers case-insensitive,
92 * don't generate the "501 unsupported" message for invalid
93 * Max-Forwards values and don't increase negative ones.
95 * Revision 1.78 2007/01/28 13:41:18 fabiankeil
96 * - Add HEAD support to finish_http_response.
97 * - Add error favicon to internal HTML error messages.
99 * Revision 1.77 2007/01/12 15:36:44 fabiankeil
100 * Mark *csp as immutable for is_untrusted_url()
101 * and is_imageurl(). Closes FR 1237736.
103 * Revision 1.76 2007/01/01 19:36:37 fabiankeil
104 * Integrate a modified version of Wil Mahan's
105 * zlib patch (PR #895531).
107 * Revision 1.75 2006/12/29 18:30:46 fabiankeil
108 * Fixed gcc43 conversion warnings,
109 * changed sprintf calls to snprintf.
111 * Revision 1.74 2006/12/24 17:37:38 fabiankeil
112 * Adjust comment in pcrs_filter_response()
113 * to recent pcrs changes. Hohoho.
115 * Revision 1.73 2006/12/23 16:01:02 fabiankeil
116 * Don't crash if pcre returns an error code
117 * that pcrs didn't expect. Fixes BR 1621173.
119 * Revision 1.72 2006/12/22 18:52:53 fabiankeil
120 * Modified is_untrusted_url to complain in case of
121 * write errors and to give a reason when adding new
122 * entries to the trustfile. Closes FR 1097611.
124 * Revision 1.71 2006/12/22 14:24:52 fabiankeil
125 * Skip empty filter files in pcrs_filter_response,
126 * but don't ignore the ones that come afterwards.
127 * Fixes parts of BR 1619208.
129 * Revision 1.70 2006/12/09 13:33:15 fabiankeil
130 * Added some sanity checks for get_last_url().
131 * Fixed possible segfault caused by my last commit.
133 * Revision 1.69 2006/12/08 12:39:13 fabiankeil
134 * Let get_last_url() catch https URLs as well.
136 * Revision 1.68 2006/12/05 14:45:48 fabiankeil
137 * Make sure get_last_url() behaves like advertised
138 * and fast-redirects{} can be combined with redirect{}.
140 * Revision 1.67 2006/11/28 15:19:43 fabiankeil
141 * Implemented +redirect{s@foo@bar@} to generate
142 * a redirect based on a rewritten version of the
145 * Revision 1.66 2006/09/23 13:26:38 roro
146 * Replace TABs by spaces in source code.
148 * Revision 1.65 2006/09/21 12:54:43 fabiankeil
149 * Fix +redirect{}. Didn't work with -fast-redirects.
151 * Revision 1.64 2006/08/31 10:55:49 fabiankeil
152 * Block requests for untrusted URLs with status
153 * code 403 instead of 200.
155 * Revision 1.63 2006/08/31 10:11:28 fabiankeil
156 * Don't free p which is still in use and will be later
157 * freed by free_map(). Don't claim the referrer is unknown
158 * when the client didn't set one.
160 * Revision 1.62 2006/08/14 00:27:47 david__schmidt
161 * Feature request 595948: Re-Filter logging in single line
163 * Revision 1.61 2006/08/03 02:46:41 david__schmidt
164 * Incorporate Fabian Keil's patch work:
\rhttp://www.fabiankeil.de/sourcecode/privoxy/
166 * Revision 1.60 2006/07/18 14:48:46 david__schmidt
167 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
168 * with what was really the latest development (the v_3_0_branch branch)
170 * Revision 1.58.2.9 2006/01/29 23:10:56 david__schmidt
171 * Multiple filter file support
173 * Revision 1.58.2.8 2005/05/07 21:50:55 david__schmidt
174 * A few memory leaks plugged (mostly on error paths)
176 * Revision 1.58.2.7 2004/10/03 12:53:32 david__schmidt
177 * Add the ability to check jpeg images for invalid
178 * lengths of comment blocks. Defensive strategy
179 * against the exploit:
180 * Microsoft Security Bulletin MS04-028
181 * Buffer Overrun in JPEG Processing (GDI+) Could
182 * Allow Code Execution (833987)
183 * Enabled with +inspect-jpegs in actions files.
185 * Revision 1.58.2.6 2003/12/06 22:18:27 gliptak
186 * Correcting compile problem with FEATURE_IMAGE_BLOCKING
188 * Revision 1.58.2.5 2003/11/11 13:10:31 oes
189 * Fixed bug #839859: "See why" link URL now gets url-encoded.
191 * Revision 1.58.2.4 2003/02/28 12:52:45 oes
194 * Revision 1.58.2.3 2002/09/25 14:51:51 oes
195 * Added basic support for OPTIONS and TRACE HTTP methods:
196 * New function direct_response which handles OPTIONS and
197 * TRACE requests whose Max-Forwards header field is zero.
199 * Revision 1.58.2.2 2002/08/01 17:18:28 oes
200 * Fixed BR 537651 / SR 579724 (MSIE image detect improper for IE/Mac)
202 * Revision 1.58.2.1 2002/07/26 15:18:53 oes
203 * - Bugfix: Executing a filters without jobs no longer results in
204 * turing off *all* filters.
205 * - Security fix: Malicious web servers can't cause a seg fault
206 * through bogus chunk sizes anymore
208 * Revision 1.58 2002/04/24 02:11:17 oes
209 * Jon's multiple AF patch: url_actions now evaluates rules
212 * Revision 1.57 2002/04/08 20:38:34 swa
215 * Revision 1.56 2002/04/05 15:51:24 oes
216 * - bugfix: error-pages now get correct request protocol
217 * - fix for invalid HTML in trust info
219 * Revision 1.55 2002/04/02 16:13:51 oes
220 * Fix: No "Go there anyway" for SSL
222 * Revision 1.54 2002/04/02 14:55:56 oes
223 * Bugfix: is_untrusted_url() now depends on FEATURE_TRUST, not FEATURE_COOKIE_JAR
225 * Revision 1.53 2002/03/26 22:29:54 swa
226 * we have a new homepage!
228 * Revision 1.52 2002/03/24 16:35:57 jongfoster
231 * Revision 1.51 2002/03/24 15:23:33 jongfoster
234 * Revision 1.50 2002/03/24 13:25:43 swa
235 * name change related issues
237 * Revision 1.49 2002/03/16 20:29:14 oes
240 * Revision 1.48 2002/03/13 20:25:34 oes
241 * Better logging for content filters
243 * Revision 1.47 2002/03/13 00:30:52 jongfoster
245 * Added option of always sending redirect for imageblock,
246 * currently disabled with #if 0.
248 * Revision 1.46 2002/03/12 01:42:49 oes
249 * Introduced modular filters
251 * Revision 1.45 2002/03/08 16:47:50 oes
252 * Added choice beween GIF and PNG built-in images
254 * Revision 1.44 2002/03/07 03:49:31 oes
255 * - Fixed compiler warnings etc
256 * - Changed built-in images from GIF to PNG
257 * (with regard to Unisys patent issue)
258 * - Added a 4x4 pattern PNG which is less intrusive
259 * than the logo but also clearly marks the deleted banners
261 * Revision 1.43 2002/01/22 23:51:59 jongfoster
262 * Replacing strsav() with the safer string_append().
264 * Adding missing html_encode() to error message generators. Where encoded
265 * and unencoded versions of a string were provided, removing the unencoded
268 * Revision 1.42 2002/01/17 21:00:32 jongfoster
269 * Moving all our URL and URL pattern parsing code to urlmatch.c.
271 * Using a single, simple url_match(pattern,url) function - rather than
272 * the 3-line match routine which was repeated all over the place.
274 * Renaming free_url to free_url_spec, since it frees a struct url_spec.
276 * Using parse_http_url() to parse URLs without faking a HTTP
277 * request line for parse_http_request().
279 * Revision 1.41 2001/11/13 00:14:07 jongfoster
280 * Fixing stupid bug now I've figured out what || means.
281 * (It always returns 0 or 1, not one of it's paramaters.)
283 * Revision 1.40 2001/10/26 17:37:55 oes
284 * - Re-enabled Netscape 200/404 bug workaround in block_url():
285 * - Removed OS/2 special case
286 * - Made block_url() independant from sed() having been run
287 * - Made trust_url independant from sed() having been run
288 * - Made is_imageurl independant from sed() having been run.
289 * It now checks User-Agent: and Accept: by itself.
292 * Revision 1.39 2001/10/25 03:40:48 david__schmidt
293 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
294 * threads to call select() simultaneously. So, it's time to do a real, live,
295 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
296 * (native). Both versions will work, but using __OS2__ offers multi-threading.
298 * Revision 1.38 2001/10/23 21:32:33 jongfoster
299 * Adding error-checking to selected functions
301 * Revision 1.37 2001/10/22 15:33:56 david__schmidt
302 * Special-cased OS/2 out of the Netscape-abort-on-404-in-js problem in
303 * filters.c. Added a FIXME in front of the offending code. I'll gladly
304 * put in a better/more robust fix for all parties if one is presented...
305 * It seems that just returning 200 instead of 404 would pretty much fix
306 * it for everyone, but I don't know all the history of the problem.
308 * Revision 1.36 2001/10/10 16:44:16 oes
309 * Added match_portlist function
311 * Revision 1.35 2001/10/07 15:41:23 oes
312 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
314 * New function remove_chunked_transfer_coding that strips chunked
315 * transfer coding to plain and is called by pcrs_filter_response
316 * and gif_deanimate_response if neccessary
318 * Improved handling of zero-change re_filter runs
320 * pcrs_filter_response and gif_deanimate_response now remove
321 * chunked transfer codeing before processing the body.
323 * Revision 1.34 2001/09/20 15:49:36 steudten
325 * Fix BUG: Change int size to size_t size in pcrs_filter_response().
326 * See cgi.c fill_template().
328 * Revision 1.33 2001/09/16 17:05:14 jongfoster
329 * Removing unused #include showarg.h
331 * Revision 1.32 2001/09/16 13:21:27 jongfoster
332 * Changes to use new list functions.
334 * Revision 1.31 2001/09/16 11:38:02 jongfoster
335 * Splitting fill_template() into 2 functions:
336 * template_load() loads the file
337 * template_fill() performs the PCRS regexps.
338 * This is because the CGI edit interface has a "table row"
339 * template which is used many times in the page - this
340 * change means it's only loaded from disk once.
342 * Revision 1.30 2001/09/16 11:00:10 jongfoster
343 * New function alloc_http_response, for symmetry with free_http_response
345 * Revision 1.29 2001/09/13 23:32:40 jongfoster
346 * Moving image data to cgi.c rather than cgi.h
347 * Fixing a GPF under Win32 (and any other OS that protects global
348 * constants from being written to).
350 * Revision 1.28 2001/09/10 10:18:51 oes
351 * Silenced compiler warnings
353 * Revision 1.27 2001/08/05 16:06:20 jongfoster
354 * Modifiying "struct map" so that there are now separate header and
355 * "map_entry" structures. This means that functions which modify a
356 * map no longer need to return a pointer to the modified map.
357 * Also, it no longer reverses the order of the entries (which may be
358 * important with some advanced template substitutions).
360 * Revision 1.26 2001/07/30 22:08:36 jongfoster
361 * Tidying up #defines:
362 * - All feature #defines are now of the form FEATURE_xxx
363 * - Permanently turned off WIN_GUI_EDIT
364 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
366 * Revision 1.25 2001/07/26 10:09:46 oes
367 * Made browser detection a little less naive
369 * Revision 1.24 2001/07/25 17:22:51 oes
370 * Added workaround for Netscape bug that prevents display of page when loading a component fails.
372 * Revision 1.23 2001/07/23 13:40:12 oes
373 * Fixed bug that caused document body to be dropped when pcrs joblist was empty.
375 * Revision 1.22 2001/07/18 12:29:34 oes
376 * - Made gif_deanimate_response respect
377 * csp->action->string[ACTION_STRING_DEANIMATE]
378 * - Logging cosmetics
380 * Revision 1.21 2001/07/13 13:59:53 oes
381 * - Introduced gif_deanimate_response which shares the
382 * generic content modification interface of pcrs_filter_response
383 * and acts as a wrapper to deanimate.c:gif_deanimate()
384 * - Renamed re_process_buffer to pcrs_filter_response
385 * - pcrs_filter_response now returns NULL on failiure
386 * - Removed all #ifdef PCRS
388 * Revision 1.20 2001/07/01 17:01:04 oes
389 * Added comments and missing return statement in is_untrusted_url()
391 * Revision 1.19 2001/06/29 21:45:41 oes
392 * Indentation, CRLF->LF, Tab-> Space
394 * Revision 1.18 2001/06/29 13:27:38 oes
395 * - Cleaned up, renamed and reorderd functions
396 * and improved comments
399 * - Ported to CGI platform. Now delivers
400 * http_response or NULL
401 * - Unified HTML and GIF generation (moved image detection
402 * and GIF generation here from jcc.c:chat())
403 * - Fixed HTTP status to:
404 * - 403 (Forbidden) for the "blocked" HTML message
405 * - 200 (OK) for GIF answers
406 * - 302 (Redirect) for redirect to GIF
409 * - Ported to CGI platform. Now delivers
410 * http_response or NULL
411 * - Separated detection of untrusted URL into
412 * (bool)is_untrusted_url
413 * - Added enforcement of untrusted requests
415 * - Moved redirect_url() from cgi.c to here
416 * and ported it to the CGI platform
418 * - Removed logentry from cancelled commit
420 * Revision 1.17 2001/06/09 10:55:28 jongfoster
421 * Changing BUFSIZ ==> BUFFER_SIZE
423 * Revision 1.16 2001/06/07 23:10:26 jongfoster
424 * Allowing unanchored domain patterns to back off and retry
425 * if they partially match. Optimized right-anchored patterns.
426 * Moving ACL and forward files into config file.
427 * Replacing struct gateway with struct forward_spec
429 * Revision 1.15 2001/06/03 19:12:00 oes
430 * extracted-CGI relevant stuff
432 * Revision 1.14 2001/06/01 10:30:55 oes
433 * Added optional left-anchoring to domaincmp
435 * Revision 1.13 2001/05/31 21:21:30 jongfoster
436 * Permissionsfile / actions file changes:
437 * - Changed "permission" to "action" throughout
438 * - changes to file format to allow string parameters
439 * - Moved helper functions to actions.c
441 * Revision 1.12 2001/05/31 17:35:20 oes
443 * - Enhanced domain part globbing with infix and prefix asterisk
444 * matching and optional unanchored operation
446 * Revision 1.11 2001/05/29 11:53:23 oes
447 * "See why" link added to "blocked" page
449 * Revision 1.10 2001/05/29 09:50:24 jongfoster
450 * Unified blocklist/imagelist/permissionslist.
451 * File format is still under discussion, but the internal changes
454 * Also modified interceptor behaviour:
455 * - We now intercept all URLs beginning with one of the following
456 * prefixes (and *only* these prefixes):
458 * * http://ijbswa.sf.net/config/
459 * * http://ijbswa.sourceforge.net/config/
460 * - New interceptors "home page" - go to http://i.j.b/ to see it.
461 * - Internal changes so that intercepted and fast redirect pages
462 * are not replaced with an image.
463 * - Interceptors now have the option to send a binary page direct
464 * to the client. (i.e. ijb-send-banner uses this)
465 * - Implemented show-url-info interceptor. (Which is why I needed
466 * the above interceptors changes - a typical URL is
467 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
468 * The previous mechanism would not have intercepted that, and
469 * if it had been intercepted then it then it would have replaced
472 * Revision 1.9 2001/05/27 22:17:04 oes
474 * - re_process_buffer no longer writes the modified buffer
475 * to the client, which was very ugly. It now returns the
476 * buffer, which it is then written by chat.
478 * - content_length now adjusts the Content-Length: header
479 * for modified documents rather than crunch()ing it.
480 * (Length info in csp->content_length, which is 0 for
481 * unmodified documents)
483 * - For this to work, sed() is called twice when filtering.
485 * Revision 1.8 2001/05/26 17:13:28 jongfoster
486 * Filled in a function comment.
488 * Revision 1.7 2001/05/26 15:26:15 jongfoster
489 * ACL feature now provides more security by immediately dropping
490 * connections from untrusted hosts.
492 * Revision 1.6 2001/05/26 00:28:36 jongfoster
493 * Automatic reloading of config file.
494 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
495 * Most of the global variables have been moved to a new
496 * struct configuration_spec, accessed through csp->config->globalname
497 * Most of the globals remaining are used by the Win32 GUI.
499 * Revision 1.5 2001/05/25 22:34:30 jongfoster
502 * Revision 1.4 2001/05/22 18:46:04 oes
504 * - Enabled filtering banners by size rather than URL
505 * by adding patterns that replace all standard banner
506 * sizes with the "Junkbuster" gif to the re_filterfile
508 * - Enabled filtering WebBugs by providing a pattern
509 * which kills all 1x1 images
511 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
512 * which is selected by the (nonstandard and therefore
513 * capital) letter 'U' in the option string.
514 * It causes the quantifiers to be ungreedy by default.
515 * Appending a ? turns back to greedy (!).
517 * - Added a new interceptor ijb-send-banner, which
518 * sends back the "Junkbuster" gif. Without imagelist or
519 * MSIE detection support, or if tinygif = 1, or the
520 * URL isn't recognized as an imageurl, a lame HTML
521 * explanation is sent instead.
523 * - Added new feature, which permits blocking remote
524 * script redirects and firing back a local redirect
526 * The feature is conditionally compiled, i.e. it
527 * can be disabled with --disable-fast-redirects,
528 * plus it must be activated by a "fast-redirects"
529 * line in the config file, has its own log level
530 * and of course wants to be displayed by show-proxy-args
531 * Note: Boy, all the #ifdefs in 1001 locations and
532 * all the fumbling with configure.in and acconfig.h
533 * were *way* more work than the feature itself :-(
535 * - Because a generic redirect template was needed for
536 * this, tinygif = 3 now uses the same.
538 * - Moved GIFs, and other static HTTP response templates
543 * - Removed some >400 CRs again (Jon, you really worked
546 * Revision 1.3 2001/05/20 16:44:47 jongfoster
547 * Removing last hardcoded Junkbusters.com URLs.
549 * Revision 1.2 2001/05/20 01:21:20 jongfoster
550 * Version 2.9.4 checkin.
551 * - Merged popupfile and cookiefile, and added control over PCRS
552 * filtering, in new "permissionsfile".
553 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
554 * file error you now get a message box (in the Win32 GUI) rather
555 * than the program exiting with no explanation.
556 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
558 * - Removed tabs from "config"
559 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
560 * - Bumped up version number.
562 * Revision 1.1.1.1 2001/05/15 13:58:52 oes
563 * Initial import of version 2.9.3 source tree
566 *********************************************************************/
572 #include <sys/types.h>
581 #endif /* ndef __OS2__ */
582 #include <netinet/in.h>
584 #include <winsock2.h>
585 #endif /* ndef _WIN32 */
589 #endif /* def __OS2__ */
597 #include "jbsockets.h"
598 #include "miscutil.h"
602 #include "deanimate.h"
603 #include "urlmatch.h"
610 const char filters_h_rcs[] = FILTERS_H_VERSION;
612 /* Fix a problem with Solaris. There should be no effect on other
614 * Solaris's isspace() is a macro which uses it's argument directly
615 * as an array index. Therefore we need to make sure that high-bit
616 * characters generate +ve values, and ideally we also want to make
617 * the argument match the declared parameter type of "int".
619 #define ijb_isdigit(__X) isdigit((int)(unsigned char)(__X))
623 /*********************************************************************
625 * Function : block_acl
627 * Description : Block this request?
628 * Decide yes or no based on ACL file.
631 * 1 : dst = The proxy or gateway address this is going to.
632 * Or NULL to check all possible targets.
633 * 2 : csp = Current client state (buffers, headers, etc...)
634 * Also includes the client IP address.
636 * Returns : 0 = FALSE (don't block) and 1 = TRUE (do block)
638 *********************************************************************/
639 int block_acl(struct access_control_addr *dst, struct client_state *csp)
641 struct access_control_list *acl = csp->config->acl;
643 /* if not using an access control list, then permit the connection */
649 /* search the list */
652 if ((csp->ip_addr_long & acl->src->mask) == acl->src->addr)
656 /* Just want to check if they have any access */
657 if (acl->action == ACL_PERMIT)
662 else if ( ((dst->addr & acl->dst->mask) == acl->dst->addr)
663 && ((dst->port == acl->dst->port) || (acl->dst->port == 0)))
665 if (acl->action == ACL_PERMIT)
683 /*********************************************************************
685 * Function : acl_addr
687 * Description : Called from `load_config' to parse an ACL address.
690 * 1 : aspec = String specifying ACL address.
691 * 2 : aca = struct access_control_addr to fill in.
693 * Returns : 0 => Ok, everything else is an error.
695 *********************************************************************/
696 int acl_addr(const char *aspec, struct access_control_addr *aca)
701 char *acl_spec = NULL;
707 * Use a temporary acl spec copy so we can log
708 * the unmodified original in case of parse errors.
710 acl_spec = strdup(aspec);
711 if (acl_spec == NULL)
713 /* XXX: This will be logged as parse error. */
717 if ((p = strchr(acl_spec, '/')) != NULL)
720 if (ijb_isdigit(*p) == 0)
725 masklength = atoi(p);
728 if ((masklength < 0) || (masklength > 32))
734 if ((p = strchr(acl_spec, ':')) != NULL)
739 port = strtol(p, &endptr, 10);
741 if (port <= 0 || port > 65535 || *endptr != '\0')
748 aca->port = (unsigned long)port;
750 aca->addr = ntohl(resolve_hostname_to_ip(acl_spec));
753 if (aca->addr == INADDR_NONE)
755 /* XXX: This will be logged as parse error. */
759 /* build the netmask */
761 for (i=1; i <= masklength ; i++)
763 aca->mask |= (1 << (32 - i));
766 /* now mask off the host portion of the ip address
767 * (i.e. save on the network portion of the address).
769 aca->addr = aca->addr & aca->mask;
774 #endif /* def FEATURE_ACL */
777 /*********************************************************************
779 * Function : block_url
781 * Description : Called from `chat'. Check to see if we need to block this.
784 * 1 : csp = Current client state (buffers, headers, etc...)
786 * Returns : NULL => unblocked, else HTTP block response
788 *********************************************************************/
789 struct http_response *block_url(struct client_state *csp)
791 struct http_response *rsp;
792 const char *new_content_type = NULL;
795 * If it's not blocked, don't block it ;-)
797 if ((csp->action->flags & ACTION_BLOCK) == 0)
801 if (csp->action->flags & ACTION_REDIRECT)
803 log_error(LOG_LEVEL_ERROR, "redirect{} overruled by block.");
806 * Else, prepare a response
808 if (NULL == (rsp = alloc_http_response()))
810 return cgi_error_memory();
814 * If it's an image-url, send back an image or redirect
815 * as specified by the relevant +image action
817 #ifdef FEATURE_IMAGE_BLOCKING
818 if (((csp->action->flags & ACTION_IMAGE_BLOCKER) != 0)
822 /* determine HOW images should be blocked */
823 p = csp->action->string[ACTION_STRING_IMAGE_BLOCKER];
825 if(csp->action->flags & ACTION_HANDLE_AS_EMPTY_DOCUMENT)
827 log_error(LOG_LEVEL_ERROR, "handle-as-empty-document overruled by handle-as-image.");
829 #if 1 /* Two alternative strategies, use this one for now: */
831 /* and handle accordingly: */
832 if ((p == NULL) || (0 == strcmpic(p, "pattern")))
834 rsp->status = strdup("403 Request blocked by Privoxy");
835 if (rsp->status == NULL)
837 free_http_response(rsp);
838 return cgi_error_memory();
840 rsp->body = bindup(image_pattern_data, image_pattern_length);
841 if (rsp->body == NULL)
843 free_http_response(rsp);
844 return cgi_error_memory();
846 rsp->content_length = image_pattern_length;
848 if (enlist_unique_header(rsp->headers, "Content-Type", BUILTIN_IMAGE_MIMETYPE))
850 free_http_response(rsp);
851 return cgi_error_memory();
855 else if (0 == strcmpic(p, "blank"))
857 rsp->status = strdup("403 Request blocked by Privoxy");
858 if (rsp->status == NULL)
860 free_http_response(rsp);
861 return cgi_error_memory();
863 rsp->body = bindup(image_blank_data, image_blank_length);
864 if (rsp->body == NULL)
866 free_http_response(rsp);
867 return cgi_error_memory();
869 rsp->content_length = image_blank_length;
871 if (enlist_unique_header(rsp->headers, "Content-Type", BUILTIN_IMAGE_MIMETYPE))
873 free_http_response(rsp);
874 return cgi_error_memory();
880 rsp->status = strdup("302 Local Redirect from Privoxy");
881 if (rsp->status == NULL)
883 free_http_response(rsp);
884 return cgi_error_memory();
887 if (enlist_unique_header(rsp->headers, "Location", p))
889 free_http_response(rsp);
890 return cgi_error_memory();
894 #else /* Following code is disabled for now */
896 /* and handle accordingly: */
897 if ((p == NULL) || (0 == strcmpic(p, "pattern")))
899 p = CGI_PREFIX "send-banner?type=pattern";
901 else if (0 == strcmpic(p, "blank"))
903 p = CGI_PREFIX "send-banner?type=blank";
905 rsp->status = strdup("302 Local Redirect from Privoxy");
906 if (rsp->status == NULL)
908 free_http_response(rsp);
909 return cgi_error_memory();
912 if (enlist_unique_header(rsp->headers, "Location", p))
914 free_http_response(rsp);
915 return cgi_error_memory();
917 #endif /* Preceeding code is disabled for now */
919 else if(csp->action->flags & ACTION_HANDLE_AS_EMPTY_DOCUMENT)
922 * Send empty document.
924 new_content_type = csp->action->string[ACTION_STRING_CONTENT_TYPE];
927 rsp->body = strdup(" ");
928 rsp->content_length = 1;
930 rsp->status = strdup("403 Request blocked by Privoxy");
931 if (rsp->status == NULL)
933 free_http_response(rsp);
934 return cgi_error_memory();
936 if (new_content_type != 0)
938 log_error(LOG_LEVEL_HEADER, "Overwriting Content-Type with %s", new_content_type);
939 if (enlist_unique_header(rsp->headers, "Content-Type", new_content_type))
941 free_http_response(rsp);
942 return cgi_error_memory();
947 #endif /* def FEATURE_IMAGE_BLOCKING */
950 * Else, generate an HTML "blocked" message:
954 struct map * exports;
958 * Workaround for stupid Netscape bug which prevents
959 * pages from being displayed if loading a referenced
960 * JavaScript or style sheet fails. So make it appear
961 * as if it succeeded.
963 if ( NULL != (p = get_header_value(csp->headers, "User-Agent:"))
964 && !strncmpic(p, "mozilla", 7) /* Catch Netscape but */
965 && !strstr(p, "Gecko") /* save Mozilla, */
966 && !strstr(p, "compatible") /* MSIE */
967 && !strstr(p, "Opera")) /* and Opera. */
969 rsp->status = strdup("200 Request for blocked URL");
973 rsp->status = strdup("403 Request for blocked URL");
976 if (rsp->status == NULL)
978 free_http_response(rsp);
979 return cgi_error_memory();
982 exports = default_exports(csp, NULL);
985 free_http_response(rsp);
986 return cgi_error_memory();
989 #ifdef FEATURE_FORCE_LOAD
990 err = map(exports, "force-prefix", 1, FORCE_PREFIX, 1);
992 * Export the force conditional block killer if
994 * - Privoxy was compiled without FEATURE_FORCE_LOAD, or
995 * - Privoxy is configured to enforce blocks, or
996 * - it's a CONNECT request and enforcing wouldn't work anyway.
998 if ((csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
999 || (0 == strcmpic(csp->http->gpc, "connect")))
1000 #endif /* ndef FEATURE_FORCE_LOAD */
1002 err = map_block_killer(exports, "force-support");
1005 if (!err) err = map(exports, "protocol", 1, csp->http->ssl ? "https://" : "http://", 1);
1006 if (!err) err = map(exports, "hostport", 1, html_encode(csp->http->hostport), 0);
1007 if (!err) err = map(exports, "path", 1, html_encode(csp->http->path), 0);
1008 if (!err) err = map(exports, "path-ue", 1, url_encode(csp->http->path), 0);
1013 free_http_response(rsp);
1014 return cgi_error_memory();
1017 err = template_fill_for_cgi(csp, "blocked", exports, rsp);
1020 free_http_response(rsp);
1021 return cgi_error_memory();
1024 rsp->reason = RSP_REASON_BLOCKED;
1026 return finish_http_response(csp, rsp);
1031 #ifdef FEATURE_TRUST
1032 /*********************************************************************
1034 * Function : trust_url FIXME: I should be called distrust_url
1036 * Description : Calls is_untrusted_url to determine if the URL is trusted
1037 * and if not, returns a HTTP 403 response with a reject message.
1040 * 1 : csp = Current client state (buffers, headers, etc...)
1042 * Returns : NULL => trusted, else http_response.
1044 *********************************************************************/
1045 struct http_response *trust_url(struct client_state *csp)
1047 struct http_response *rsp;
1048 struct map * exports;
1049 char buf[BUFFER_SIZE];
1051 struct url_spec **tl;
1056 * Don't bother to work on trusted URLs
1058 if (!is_untrusted_url(csp))
1064 * Else, prepare a response:
1066 if (NULL == (rsp = alloc_http_response()))
1068 return cgi_error_memory();
1071 rsp->status = strdup("403 Request blocked by Privoxy");
1072 exports = default_exports(csp, NULL);
1073 if (exports == NULL || rsp->status == NULL)
1075 free_http_response(rsp);
1076 return cgi_error_memory();
1080 * Export the protocol, host, port, and referrer information
1082 err = map(exports, "hostport", 1, csp->http->hostport, 1);
1083 if (!err) err = map(exports, "protocol", 1, csp->http->ssl ? "https://" : "http://", 1);
1084 if (!err) err = map(exports, "path", 1, csp->http->path, 1);
1086 if (NULL != (p = get_header_value(csp->headers, "Referer:")))
1088 if (!err) err = map(exports, "referrer", 1, html_encode(p), 0);
1092 if (!err) err = map(exports, "referrer", 1, "none set", 1);
1098 free_http_response(rsp);
1099 return cgi_error_memory();
1103 * Export the trust list
1106 for (tl = csp->config->trust_list; (t = *tl) != NULL ; tl++)
1108 snprintf(buf, sizeof(buf), "<li>%s</li>\n", t->spec);
1109 string_append(&p, buf);
1111 err = map(exports, "trusted-referrers", 1, p, 0);
1116 free_http_response(rsp);
1117 return cgi_error_memory();
1121 * Export the trust info, if available
1123 if (csp->config->trust_info->first)
1125 struct list_entry *l;
1128 for (l = csp->config->trust_info->first; l ; l = l->next)
1130 snprintf(buf, sizeof(buf), "<li> <a href=\"%s\">%s</a><br>\n", l->str, l->str);
1131 string_append(&p, buf);
1133 err = map(exports, "trust-info", 1, p, 0);
1137 err = map_block_killer(exports, "have-trust-info");
1143 free_http_response(rsp);
1144 return cgi_error_memory();
1148 * Export the force conditional block killer if
1150 * - Privoxy was compiled without FEATURE_FORCE_LOAD, or
1151 * - Privoxy is configured to enforce blocks, or
1152 * - it's a CONNECT request and enforcing wouldn't work anyway.
1154 #ifdef FEATURE_FORCE_LOAD
1155 if ((csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1156 || (0 == strcmpic(csp->http->gpc, "connect")))
1158 err = map_block_killer(exports, "force-support");
1162 err = map(exports, "force-prefix", 1, FORCE_PREFIX, 1);
1164 #else /* ifndef FEATURE_FORCE_LOAD */
1165 err = map_block_killer(exports, "force-support");
1166 #endif /* ndef FEATURE_FORCE_LOAD */
1171 free_http_response(rsp);
1172 return cgi_error_memory();
1176 * Build the response
1178 err = template_fill_for_cgi(csp, "untrusted", exports, rsp);
1181 free_http_response(rsp);
1182 return cgi_error_memory();
1184 rsp->reason = RSP_REASON_UNTRUSTED;
1186 return finish_http_response(csp, rsp);
1188 #endif /* def FEATURE_TRUST */
1191 /*********************************************************************
1193 * Function : compile_dynamic_pcrs_job_list
1195 * Description : Compiles a dynamic pcrs job list (one with variables
1196 * resolved at request time)
1199 * 1 : csp = Current client state (buffers, headers, etc...)
1200 * 2 : b = The filter list to compile
1202 * Returns : NULL in case of errors, otherwise the
1205 *********************************************************************/
1206 pcrs_job *compile_dynamic_pcrs_job_list(const struct client_state *csp, const struct re_filterfile_spec *b)
1208 struct list_entry *pattern;
1209 pcrs_job *job_list = NULL;
1210 pcrs_job *dummy = NULL;
1211 pcrs_job *lastjob = NULL;
1214 const struct pcrs_variable variables[] =
1216 {"url", csp->http->url, 1},
1217 {"path", csp->http->path, 1},
1218 {"host", csp->http->host, 1},
1219 {"origin", csp->ip_addr_str, 1},
1223 for (pattern = b->patterns->first; pattern != NULL; pattern = pattern->next)
1225 assert(pattern->str != NULL);
1227 dummy = pcrs_compile_dynamic_command(pattern->str, variables, &error);
1231 log_error(LOG_LEVEL_ERROR,
1232 "Adding filter job \'%s\' to dynamic filter %s failed: %s",
1233 pattern->str, b->name, pcrs_strerror(error));
1238 if (error == PCRS_WARN_TRUNCATION)
1240 log_error(LOG_LEVEL_ERROR,
1241 "At least one of the variables in \'%s\' had to "
1242 "be truncated before compilation", pattern->str);
1244 if (job_list == NULL)
1250 lastjob->next = dummy;
1260 /*********************************************************************
1262 * Function : rewrite_url
1264 * Description : Rewrites a URL with a single pcrs command
1265 * and returns the result if it differs from the
1266 * original and isn't obviously invalid.
1269 * 1 : old_url = URL to rewrite.
1270 * 2 : pcrs_command = pcrs command formatted as string (s@foo@bar@)
1273 * Returns : NULL if the pcrs_command didn't change the url, or
1274 * the result of the modification.
1276 *********************************************************************/
1277 char *rewrite_url(char *old_url, const char *pcrs_command)
1279 char *new_url = NULL;
1283 assert(pcrs_command);
1285 new_url = pcrs_execute_single_command(old_url, pcrs_command, &hits);
1289 log_error(LOG_LEVEL_REDIRECTS,
1290 "pcrs command \"%s\" didn't change \"%s\".",
1291 pcrs_command, old_url);
1296 log_error(LOG_LEVEL_REDIRECTS,
1297 "executing pcrs command \"%s\" to rewrite %s failed: %s",
1298 pcrs_command, old_url, pcrs_strerror(hits));
1301 else if (strncmpic(new_url, "http://", 7) && strncmpic(new_url, "https://", 8))
1303 log_error(LOG_LEVEL_ERROR,
1304 "pcrs command \"%s\" changed \"%s\" to \"%s\" (%u hi%s), "
1305 "but the result doesn't look like a valid URL and will be ignored.",
1306 pcrs_command, old_url, new_url, hits, (hits == 1) ? "t" : "ts");
1311 log_error(LOG_LEVEL_REDIRECTS,
1312 "pcrs command \"%s\" changed \"%s\" to \"%s\" (%u hi%s).",
1313 pcrs_command, old_url, new_url, hits, (hits == 1) ? "t" : "ts");
1321 #ifdef FEATURE_FAST_REDIRECTS
1322 /*********************************************************************
1324 * Function : get_last_url
1326 * Description : Search for the last URL inside a string.
1327 * If the string already is a URL, it will
1328 * be the first URL found.
1331 * 1 : subject = the string to check
1332 * 2 : redirect_mode = +fast-redirect{} mode
1334 * Returns : NULL if no URL was found, or
1335 * the last URL found.
1337 *********************************************************************/
1338 char *get_last_url(char *subject, const char *redirect_mode)
1340 char *new_url = NULL;
1344 assert(redirect_mode);
1346 subject = strdup(subject);
1347 if (subject == NULL)
1349 log_error(LOG_LEVEL_ERROR, "Out of memory while searching for redirects.");
1353 if (0 == strcmpic(redirect_mode, "check-decoded-url"))
1355 log_error(LOG_LEVEL_REDIRECTS, "Decoding \"%s\" if necessary.", subject);
1356 new_url = url_decode(subject);
1357 if (new_url != NULL)
1364 log_error(LOG_LEVEL_ERROR, "Unable to decode \"%s\".", subject);
1368 log_error(LOG_LEVEL_REDIRECTS, "Checking \"%s\" for redirects.", subject);
1371 * Find the last URL encoded in the request
1374 while ((tmp = strstr(tmp, "http://")) != NULL)
1378 tmp = (new_url != NULL) ? new_url : subject;
1379 while ((tmp = strstr(tmp, "https://")) != NULL)
1384 if ((new_url != NULL)
1385 && ( (new_url != subject)
1386 || (0 == strncmpic(subject, "http://", 7))
1387 || (0 == strncmpic(subject, "https://", 8))
1391 * Return new URL if we found a redirect
1392 * or if the subject already was a URL.
1394 * The second case makes sure that we can
1395 * chain get_last_url after another redirection check
1396 * (like rewrite_url) without losing earlier redirects.
1398 new_url = strdup(new_url);
1407 #endif /* def FEATURE_FAST_REDIRECTS */
1410 /*********************************************************************
1412 * Function : redirect_url
1414 * Description : Checks if Privoxy should answer the request with
1415 * a HTTP redirect and generates the redirect if
1419 * 1 : csp = Current client state (buffers, headers, etc...)
1421 * Returns : NULL if the request can pass, HTTP redirect otherwise.
1423 *********************************************************************/
1424 struct http_response *redirect_url(struct client_state *csp)
1426 struct http_response *rsp;
1427 #ifdef FEATURE_FAST_REDIRECTS
1429 * XXX: Do we still need FEATURE_FAST_REDIRECTS
1430 * as compile-time option? The user can easily disable
1431 * it in his action file.
1433 char * redirect_mode;
1434 #endif /* def FEATURE_FAST_REDIRECTS */
1435 char *old_url = NULL;
1436 char *new_url = NULL;
1437 char *redirection_string;
1439 if ((csp->action->flags & ACTION_REDIRECT))
1441 redirection_string = csp->action->string[ACTION_STRING_REDIRECT];
1444 * If the redirection string begins with 's',
1445 * assume it's a pcrs command, otherwise treat it as
1446 * properly formatted URL and use it for the redirection
1449 * According to RFC 2616 section 14.30 the URL
1450 * has to be absolute and if the user tries:
1451 * +redirect{shit/this/will/be/parsed/as/pcrs_command.html}
1452 * she would get undefined results anyway.
1456 if (*redirection_string == 's')
1458 old_url = csp->http->url;
1459 new_url = rewrite_url(old_url, redirection_string);
1463 log_error(LOG_LEVEL_REDIRECTS,
1464 "No pcrs command recognized, assuming that \"%s\" is already properly formatted.",
1465 redirection_string);
1466 new_url = strdup(redirection_string);
1470 #ifdef FEATURE_FAST_REDIRECTS
1471 if ((csp->action->flags & ACTION_FAST_REDIRECTS))
1473 redirect_mode = csp->action->string[ACTION_STRING_FAST_REDIRECTS];
1476 * If it exists, use the previously rewritten URL as input
1477 * otherwise just use the old path.
1479 old_url = (new_url != NULL) ? new_url : strdup(csp->http->path);
1480 new_url = get_last_url(old_url, redirect_mode);
1485 * Disable redirect checkers, so that they
1486 * will be only run more than once if the user
1487 * also enables them through tags.
1489 * From a performance point of view
1490 * it doesn't matter, but the duplicated
1491 * log messages are annoying.
1493 csp->action->flags &= ~ACTION_FAST_REDIRECTS;
1494 #endif /* def FEATURE_FAST_REDIRECTS */
1495 csp->action->flags &= ~ACTION_REDIRECT;
1497 /* Did any redirect action trigger? */
1500 if (0 == strcmpic(new_url, csp->http->url))
1502 log_error(LOG_LEVEL_ERROR,
1503 "New URL \"%s\" and old URL \"%s\" are the same. Redirection loop prevented.",
1504 csp->http->url, new_url);
1509 log_error(LOG_LEVEL_REDIRECTS, "New URL is: %s", new_url);
1511 if (NULL == (rsp = alloc_http_response()))
1514 return cgi_error_memory();
1517 if ( enlist_unique_header(rsp->headers, "Location", new_url)
1518 || (NULL == (rsp->status = strdup("302 Local Redirect from Privoxy"))) )
1521 free_http_response(rsp);
1522 return cgi_error_memory();
1524 rsp->reason = RSP_REASON_REDIRECTED;
1527 return finish_http_response(csp, rsp);
1531 /* Only reached if no redirect is required */
1537 #ifdef FEATURE_IMAGE_BLOCKING
1538 /*********************************************************************
1540 * Function : is_imageurl
1542 * Description : Given a URL, decide whether it is an image or not,
1543 * using either the info from a previous +image action
1544 * or, #ifdef FEATURE_IMAGE_DETECT_MSIE, and the browser
1545 * is MSIE and not on a Mac, tell from the browser's accept
1549 * 1 : csp = Current client state (buffers, headers, etc...)
1551 * Returns : True (nonzero) if URL is an image, false (0)
1554 *********************************************************************/
1555 int is_imageurl(const struct client_state *csp)
1557 #ifdef FEATURE_IMAGE_DETECT_MSIE
1560 tmp = get_header_value(csp->headers, "User-Agent:");
1561 if (tmp && strstr(tmp, "MSIE") && !strstr(tmp, "Mac_"))
1563 tmp = get_header_value(csp->headers, "Accept:");
1564 if (tmp && strstr(tmp, "image/gif"))
1566 /* Client will accept HTML. If this seems counterintuitive,
1576 #endif /* def FEATURE_IMAGE_DETECT_MSIE */
1578 return ((csp->action->flags & ACTION_IMAGE) != 0);
1581 #endif /* def FEATURE_IMAGE_BLOCKING */
1584 #ifdef FEATURE_TRUST
1585 /*********************************************************************
1587 * Function : is_untrusted_url
1589 * Description : Should we "distrust" this URL (and block it)?
1591 * Yes if it matches a line in the trustfile, or if the
1592 * referrer matches a line starting with "+" in the
1597 * 1 : csp = Current client state (buffers, headers, etc...)
1599 * Returns : 0 => trusted, 1 => untrusted
1601 *********************************************************************/
1602 int is_untrusted_url(const struct client_state *csp)
1604 struct file_list *fl;
1605 struct block_spec *b;
1606 struct url_spec **trusted_url;
1607 struct http_request rhttp[1];
1608 const char * referer;
1612 * If we don't have a trustlist, we trust everybody
1614 if (((fl = csp->tlist) == NULL) || ((b = fl->f) == NULL))
1619 memset(rhttp, '\0', sizeof(*rhttp));
1622 * Do we trust the request URL itself?
1624 for (b = b->next; b ; b = b->next)
1626 if (url_match(b->url, csp->http))
1632 if (NULL == (referer = get_header_value(csp->headers, "Referer:")))
1634 /* no referrer was supplied */
1640 * If not, do we maybe trust its referrer?
1642 err = parse_http_url(referer, rhttp, csp);
1648 for (trusted_url = csp->config->trust_list; *trusted_url != NULL; trusted_url++)
1650 if (url_match(*trusted_url, rhttp))
1652 /* if the URL's referrer is from a trusted referrer, then
1653 * add the target spec to the trustfile as an unblocked
1654 * domain and return 0 (which means it's OK).
1659 if (NULL != (fp = fopen(csp->config->trustfile, "a")))
1663 char * new_entry = strdup("~");
1665 string_append(&new_entry, csp->http->hostport);
1667 path = csp->http->path;
1668 if ( (path[0] == '/')
1670 && ((path_end = strchr(path + 2, '/')) != NULL))
1672 /* since this path points into a user's home space
1673 * be sure to include this spec in the trustfile.
1675 int path_len = path_end - path; /* save offset */
1676 path = strdup(path); /* Copy string */
1679 path_end = path + path_len; /* regenerate ptr to new buffer */
1680 *(path_end + 1) = '\0'; /* Truncate path after '/' */
1682 string_join(&new_entry, path);
1686 * Give a reason for generating this entry.
1688 string_append(&new_entry, " # Trusted referrer was: ");
1689 string_append(&new_entry, referer);
1691 if (new_entry != NULL)
1693 if (-1 == fprintf(fp, "%s\n", new_entry))
1695 log_error(LOG_LEVEL_ERROR, "Failed to append \'%s\' to trustfile \'%s\': %E",
1696 new_entry, csp->config->trustfile);
1702 /* FIXME: No way to handle out-of memory, so mostly ignoring it */
1703 log_error(LOG_LEVEL_ERROR, "Out of memory adding pattern to trust file");
1710 log_error(LOG_LEVEL_ERROR, "Failed to append new entry for \'%s\' to trustfile \'%s\': %E",
1711 csp->http->hostport, csp->config->trustfile);
1719 #endif /* def FEATURE_TRUST */
1722 /*********************************************************************
1724 * Function : pcrs_filter_response
1726 * Description : Execute all text substitutions from all applying
1727 * +filter actions on the text buffer that's been accumulated
1728 * in csp->iob->buf. If this changes the contents, set
1729 * csp->content_length to the modified size and raise the
1730 * CSP_FLAG_MODIFIED flag.
1732 * XXX: Currently pcrs_filter_response is also responsible
1733 * for dechunking and decompressing. Both should be
1734 * done in separate functions so other content modifiers
1735 * profit as well, even if pcrs filtering is disabled.
1738 * 1 : csp = Current client state (buffers, headers, etc...)
1740 * Returns : a pointer to the (newly allocated) modified buffer.
1741 * or NULL if there were no hits or something went wrong
1743 *********************************************************************/
1744 char *pcrs_filter_response(struct client_state *csp)
1747 size_t size, prev_size;
1749 char *old = csp->iob->cur, *new = NULL;
1752 struct file_list *fl;
1753 struct re_filterfile_spec *b;
1754 struct list_entry *filtername;
1756 int i, found_filters = 0;
1761 if (csp->iob->cur >= csp->iob->eod)
1765 size = (size_t)(csp->iob->eod - csp->iob->cur);
1768 * Need to check the set of re_filterfiles...
1770 for (i = 0; i < MAX_AF_FILES; i++)
1783 if (0 == found_filters)
1785 log_error(LOG_LEVEL_ERROR, "Unable to get current state of regexp filtering.");
1790 * If the body has a "chunked" transfer-encoding,
1791 * get rid of it first, adjusting size and iob->eod
1793 if (csp->flags & CSP_FLAG_CHUNKED)
1795 log_error(LOG_LEVEL_RE_FILTER, "Need to de-chunk first");
1796 if (0 == (size = remove_chunked_transfer_coding(csp->iob->cur, size)))
1800 csp->iob->eod = csp->iob->cur + size;
1801 csp->flags |= CSP_FLAG_MODIFIED;
1806 * If the body has a compressed transfer-encoding,
1807 * uncompress it first, adjusting size and iob->eod.
1808 * Note that decompression occurs after de-chunking.
1810 if (csp->content_type & (CT_GZIP | CT_DEFLATE))
1812 /* Notice that we at least tried to decompress. */
1813 if (JB_ERR_OK != decompress_iob(csp))
1816 * We failed to decompress the data; there's no point
1817 * in continuing since we can't filter.
1819 * XXX: Actually the Accept-Encoding header may
1820 * just be incorrect in which case we could continue
1823 * Unset CT_GZIP and CT_DEFLATE to remember not
1824 * to modify the Content-Encoding header later.
1826 csp->content_type &= ~CT_GZIP;
1827 csp->content_type &= ~CT_DEFLATE;
1832 * Decompression gives us a completely new iob,
1833 * so we need to update.
1835 size = (size_t)(csp->iob->eod - csp->iob->cur);
1836 old = csp->iob->cur;
1838 csp->flags |= CSP_FLAG_MODIFIED;
1839 csp->content_type &= ~CT_TABOO;
1843 for (i = 0; i < MAX_AF_FILES; i++)
1846 if ((NULL == fl) || (NULL == fl->f))
1849 * Either there are no filter files
1850 * left, or this filter file just
1851 * contains no valid filters.
1853 * Continue to be sure we don't miss
1854 * valid filter files that are chained
1855 * after empty or invalid ones.
1860 * For all applying +filter actions, look if a filter by that
1861 * name exists and if yes, execute it's pcrs_joblist on the
1864 for (b = fl->f; b; b = b->next)
1866 if (b->type != FT_CONTENT_FILTER)
1868 /* Skip header filters */
1872 for (filtername = csp->action->multi[ACTION_MULTI_FILTER]->first;
1873 filtername ; filtername = filtername->next)
1875 if (strcmp(b->name, filtername->str) == 0)
1877 int current_hits = 0; /* Number of hits caused by this filter */
1878 int job_number = 0; /* Which job we're currently executing */
1879 int job_hits = 0; /* How many hits the current job caused */
1880 pcrs_job *joblist = b->joblist;
1882 if (b->dynamic) joblist = compile_dynamic_pcrs_job_list(csp, b);
1884 if (NULL == joblist)
1886 log_error(LOG_LEVEL_RE_FILTER, "Filter %s has empty joblist. Nothing to do.", b->name);
1891 /* Apply all jobs from the joblist */
1892 for (job = joblist; NULL != job; job = job->next)
1895 job_hits = pcrs_execute(job, old, size, &new, &size);
1900 * That went well. Continue filtering
1901 * and use the result of this job as
1902 * input for the next one.
1904 current_hits += job_hits;
1905 if (old != csp->iob->cur)
1914 * This job caused an unexpected error. Inform the user
1915 * and skip the rest of the jobs in this filter. We could
1916 * continue with the next job, but usually the jobs
1917 * depend on each other or are similar enough to
1918 * fail for the same reason.
1920 * At the moment our pcrs expects the error codes of pcre 3.4,
1921 * but newer pcre versions can return additional error codes.
1922 * As a result pcrs_strerror()'s error message might be
1923 * "Unknown error ...", therefore we print the numerical value
1926 * XXX: Is this important enough for LOG_LEVEL_ERROR or
1927 * should we use LOG_LEVEL_RE_FILTER instead?
1929 log_error(LOG_LEVEL_ERROR, "Skipped filter \'%s\' after job number %u: %s (%d)",
1930 b->name, job_number, pcrs_strerror(job_hits), job_hits);
1935 if (b->dynamic) pcrs_free_joblist(joblist);
1937 log_error(LOG_LEVEL_RE_FILTER,
1938 "filtering %s%s (size %d) with \'%s\' produced %d hits (new size %d).",
1939 csp->http->hostport, csp->http->path, prev_size, b->name, current_hits, size);
1941 hits += current_hits;
1948 * If there were no hits, destroy our copy and let
1949 * chat() use the original in csp->iob
1957 csp->flags |= CSP_FLAG_MODIFIED;
1958 csp->content_length = size;
1966 /*********************************************************************
1968 * Function : gif_deanimate_response
1970 * Description : Deanimate the GIF image that has been accumulated in
1971 * csp->iob->buf, set csp->content_length to the modified
1972 * size and raise the CSP_FLAG_MODIFIED flag.
1975 * 1 : csp = Current client state (buffers, headers, etc...)
1977 * Returns : a pointer to the (newly allocated) modified buffer.
1978 * or NULL in case something went wrong.
1980 *********************************************************************/
1981 char *gif_deanimate_response(struct client_state *csp)
1983 struct binbuffer *in, *out;
1987 size = (size_t)(csp->iob->eod - csp->iob->cur);
1990 * If the body has a "chunked" transfer-encoding,
1991 * get rid of it first, adjusting size and iob->eod
1993 if (csp->flags & CSP_FLAG_CHUNKED)
1995 log_error(LOG_LEVEL_DEANIMATE, "Need to de-chunk first");
1996 if (0 == (size = remove_chunked_transfer_coding(csp->iob->cur, size)))
2000 csp->iob->eod = csp->iob->cur + size;
2001 csp->flags |= CSP_FLAG_MODIFIED;
2004 if ( (NULL == (in = (struct binbuffer *)zalloc(sizeof *in )))
2005 || (NULL == (out = (struct binbuffer *)zalloc(sizeof *out))) )
2007 log_error(LOG_LEVEL_DEANIMATE, "failed! (no mem)");
2011 in->buffer = csp->iob->cur;
2014 if (gif_deanimate(in, out, strncmp("last", csp->action->string[ACTION_STRING_DEANIMATE], 4)))
2016 log_error(LOG_LEVEL_DEANIMATE, "failed! (gif parsing)");
2023 if ((int)size == out->offset)
2025 log_error(LOG_LEVEL_DEANIMATE, "GIF not changed.");
2029 log_error(LOG_LEVEL_DEANIMATE, "Success! GIF shrunk from %d bytes to %d.", size, out->offset);
2031 csp->content_length = out->offset;
2032 csp->flags |= CSP_FLAG_MODIFIED;
2042 /*********************************************************************
2044 * Function : jpeg_inspect_response
2049 * 1 : csp = Current client state (buffers, headers, etc...)
2051 * Returns : a pointer to the (newly allocated) modified buffer
2052 * or NULL in case something went wrong.
2054 *********************************************************************/
2055 char *jpeg_inspect_response(struct client_state *csp)
2057 struct binbuffer *in = NULL;
2058 struct binbuffer *out = NULL;
2062 size = (size_t)(csp->iob->eod - csp->iob->cur);
2065 * If the body has a "chunked" transfer-encoding,
2066 * get rid of it first, adjusting size and iob->eod
2068 if (csp->flags & CSP_FLAG_CHUNKED)
2070 log_error(LOG_LEVEL_DEANIMATE, "Need to de-chunk first");
2071 if (0 == (size = remove_chunked_transfer_coding(csp->iob->cur, size)))
2075 csp->iob->eod = csp->iob->cur + size;
2076 csp->flags |= CSP_FLAG_MODIFIED;
2079 if (NULL == (in = (struct binbuffer *)zalloc(sizeof *in )))
2081 log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg no mem 1)");
2085 if (NULL == (out = (struct binbuffer *)zalloc(sizeof *out)))
2087 log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg no mem 2)");
2091 in->buffer = csp->iob->cur;
2095 * Calling jpeg_inspect has the side-effect of creating and
2096 * modifying the image buffer of "out" directly.
2098 if (jpeg_inspect(in, out))
2100 log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg parsing)");
2108 csp->content_length = out->offset;
2109 csp->flags |= CSP_FLAG_MODIFIED;
2119 /*********************************************************************
2121 * Function : remove_chunked_transfer_coding
2123 * Description : In-situ remove the "chunked" transfer coding as defined
2124 * in rfc2616 from a buffer.
2127 * 1 : buffer = Pointer to the text buffer
2128 * 2 : size = Number of bytes to be processed
2130 * Returns : The new size, i.e. the number of bytes from buffer which
2131 * are occupied by the stripped body, or 0 in case something
2134 *********************************************************************/
2135 size_t remove_chunked_transfer_coding(char *buffer, const size_t size)
2138 unsigned int chunksize = 0;
2139 char *from_p, *to_p;
2142 from_p = to_p = buffer;
2144 if (sscanf(buffer, "%x", &chunksize) != 1)
2146 log_error(LOG_LEVEL_ERROR, "Invalid first chunksize while stripping \"chunked\" transfer coding");
2150 while (chunksize > 0)
2152 if (NULL == (from_p = strstr(from_p, "\r\n")))
2154 log_error(LOG_LEVEL_ERROR, "Parse error while stripping \"chunked\" transfer coding");
2158 if ((newsize += chunksize) >= size)
2160 log_error(LOG_LEVEL_ERROR, "Chunksize exceeds buffer in \"chunked\" transfer coding");
2165 memmove(to_p, from_p, (size_t) chunksize);
2166 to_p = buffer + newsize;
2167 from_p += chunksize + 2;
2169 if (sscanf(from_p, "%x", &chunksize) != 1)
2171 log_error(LOG_LEVEL_ERROR, "Parse error while stripping \"chunked\" transfer coding");
2176 /* FIXME: Should this get its own loglevel? */
2177 log_error(LOG_LEVEL_RE_FILTER, "De-chunking successful. Shrunk from %d to %d", size, newsize);
2183 /*********************************************************************
2185 * Function : url_actions
2187 * Description : Gets the actions for this URL.
2190 * 1 : http = http_request request for blocked URLs
2191 * 2 : csp = Current client state (buffers, headers, etc...)
2195 *********************************************************************/
2196 void url_actions(struct http_request *http,
2197 struct client_state *csp)
2199 struct file_list *fl;
2200 struct url_actions *b;
2203 init_current_action(csp->action);
2205 for (i = 0; i < MAX_AF_FILES; i++)
2207 if (((fl = csp->actions_list[i]) == NULL) || ((b = fl->f) == NULL))
2212 apply_url_actions(csp->action, http, b);
2219 /*********************************************************************
2221 * Function : apply_url_actions
2223 * Description : Applies a list of URL actions.
2226 * 1 : action = Destination.
2227 * 2 : http = Current URL
2228 * 3 : b = list of URL actions to apply
2232 *********************************************************************/
2233 void apply_url_actions(struct current_action_spec *action,
2234 struct http_request *http,
2235 struct url_actions *b)
2239 /* Should never happen */
2243 for (b = b->next; NULL != b; b = b->next)
2245 if (url_match(b->url, http))
2247 merge_current_action(action, b->action);
2253 /*********************************************************************
2255 * Function : get_forward_override_settings
2257 * Description : Returns forward settings as specified with the
2258 * forward-override{} action. forward-override accepts
2259 * forward lines similar to the one used in the
2260 * configuration file, but without the URL pattern.
2266 * in the configuration file can be replaced with
2267 * the action section:
2269 * {+forward-override{forward .}}
2273 * 1 : csp = Current client state (buffers, headers, etc...)
2275 * Returns : Pointer to forwarding structure in case of success.
2276 * Invalid syntax is fatal.
2278 *********************************************************************/
2279 static const struct forward_spec *get_forward_override_settings(struct client_state *csp)
2281 const char *forward_override_line = csp->action->string[ACTION_STRING_FORWARD_OVERRIDE];
2282 char forward_settings[BUFFER_SIZE];
2283 char *http_parent = NULL;
2284 /* variable names were chosen for consistency reasons. */
2285 struct forward_spec *fwd = NULL;
2289 assert(csp->action->flags & ACTION_FORWARD_OVERRIDE);
2290 /* Should be enforced by load_one_actions_file() */
2291 assert(strlen(forward_override_line) < sizeof(forward_settings) - 1);
2293 /* Create a copy ssplit can modify */
2294 strlcpy(forward_settings, forward_override_line, sizeof(forward_settings));
2296 if (NULL != csp->fwd)
2299 * XXX: Currently necessary to prevent memory
2300 * leaks when the show-url-info cgi page is visited.
2302 unload_forward_spec(csp->fwd);
2306 * allocate a new forward node, valid only for
2307 * the lifetime of this request. Save its location
2308 * in csp as well, so sweep() can free it later on.
2310 fwd = csp->fwd = zalloc(sizeof(*fwd));
2313 log_error(LOG_LEVEL_FATAL,
2314 "can't allocate memory for forward-override{%s}", forward_override_line);
2315 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2318 vec_count = ssplit(forward_settings, " \t", vec, SZ(vec), 1, 1);
2319 if ((vec_count == 2) && !strcasecmp(vec[0], "forward"))
2321 fwd->type = SOCKS_NONE;
2323 /* Parse the parent HTTP proxy host:port */
2324 http_parent = vec[1];
2327 else if (vec_count == 3)
2329 char *socks_proxy = NULL;
2331 if (!strcasecmp(vec[0], "forward-socks4"))
2333 fwd->type = SOCKS_4;
2334 socks_proxy = vec[1];
2336 else if (!strcasecmp(vec[0], "forward-socks4a"))
2338 fwd->type = SOCKS_4A;
2339 socks_proxy = vec[1];
2342 if (NULL != socks_proxy)
2344 /* Parse the SOCKS proxy host[:port] */
2345 fwd->gateway_host = strdup(socks_proxy);
2347 if (NULL != (socks_proxy = strchr(fwd->gateway_host, ':')))
2349 *socks_proxy++ = '\0';
2350 fwd->gateway_port = strtol(socks_proxy, NULL, 0);
2353 if (fwd->gateway_port <= 0)
2355 fwd->gateway_port = 1080;
2358 http_parent = vec[2];
2362 if (NULL == http_parent)
2364 log_error(LOG_LEVEL_FATAL,
2365 "Invalid forward-override syntax in: %s", forward_override_line);
2366 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2369 /* Parse http forwarding settings */
2370 if (strcmp(http_parent, ".") != 0)
2372 fwd->forward_host = strdup(http_parent);
2374 if (NULL != (http_parent = strchr(fwd->forward_host, ':')))
2376 *http_parent++ = '\0';
2377 fwd->forward_port = strtol(http_parent, NULL, 0);
2380 if (fwd->forward_port <= 0)
2382 fwd->forward_port = 8000;
2386 assert (NULL != fwd);
2388 log_error(LOG_LEVEL_CONNECT,
2389 "Overriding forwarding settings based on \'%s\'", forward_override_line);
2395 /*********************************************************************
2397 * Function : forward_url
2399 * Description : Should we forward this to another proxy?
2401 * XXX: Should be changed to make use of csp->fwd.
2404 * 1 : http = http_request request for current URL
2405 * 2 : csp = Current client state (buffers, headers, etc...)
2407 * Returns : Pointer to forwarding information.
2409 *********************************************************************/
2410 const struct forward_spec * forward_url(struct http_request *http,
2411 struct client_state *csp)
2413 static const struct forward_spec fwd_default[1] = { FORWARD_SPEC_INITIALIZER };
2414 struct forward_spec *fwd = csp->config->forward;
2416 if (csp->action->flags & ACTION_FORWARD_OVERRIDE)
2418 return get_forward_override_settings(csp);
2428 if (url_match(fwd->url, http))
2439 /*********************************************************************
2441 * Function : direct_response
2443 * Description : Check if Max-Forwards == 0 for an OPTIONS or TRACE
2444 * request and if so, return a HTTP 501 to the client.
2446 * FIXME: I have a stupid name and I should handle the
2447 * requests properly. Still, what we do here is rfc-
2448 * compliant, whereas ignoring or forwarding are not.
2451 * 1 : csp = Current client state (buffers, headers, etc...)
2453 * Returns : http_response if , NULL if nonmatch or handler fail
2455 *********************************************************************/
2456 struct http_response *direct_response(struct client_state *csp)
2458 struct http_response *rsp;
2459 struct list_entry *p;
2461 if ((0 == strcmpic(csp->http->gpc, "trace"))
2462 || (0 == strcmpic(csp->http->gpc, "options")))
2464 for (p = csp->headers->first; (p != NULL) ; p = p->next)
2466 if (!strncmpic("Max-Forwards:", p->str, 13))
2468 unsigned int max_forwards;
2471 * If it's a Max-Forwards value of zero,
2472 * we have to intercept the request.
2474 if (1 == sscanf(p->str+12, ": %u", &max_forwards) && max_forwards == 0)
2477 * FIXME: We could handle at least TRACE here,
2478 * but that would require a verbatim copy of
2479 * the request which we don't have anymore
2481 log_error(LOG_LEVEL_HEADER,
2482 "Detected header \'%s\' in OPTIONS or TRACE request. Returning 501.",
2485 /* Get mem for response or fail*/
2486 if (NULL == (rsp = alloc_http_response()))
2488 return cgi_error_memory();
2491 if (NULL == (rsp->status = strdup("501 Not Implemented")))
2493 free_http_response(rsp);
2494 return cgi_error_memory();
2498 rsp->reason = RSP_REASON_UNSUPPORTED;
2500 return(finish_http_response(csp, rsp));